Mac Schwager

Simon Le Cleac'h, Hong-Xing Yu, Michelle Guo et al. · stanford

We present a differentiable pipeline for simulating the motion of objects that represent their geometry as a continuous density field parameterized as a deep network. This includes Neural Radiance Fields (NeRFs), and other related models. From the density field, we estimate the dynamical properties of the object, including its mass, center of mass, and inertia matrix. We then introduce a differentiable contact model based on the density field for computing normal and friction forces resulting from collisions. This allows a robot to autonomously build object models that are visually and \emph{dynamically} accurate from still images and videos of objects in motion. The resulting Dynamics-Augmented Neural Objects (DANOs) are simulated with an existing differentiable simulation engine, Dojo, interacting with other standard simulation objects, such as spheres, planes, and robots specified as URDFs. A robot can use this simulation to optimize grasps and manipulation trajectories of neural objects, or to improve the neural object models through gradient-based real-to-simulation transfer. We demonstrate the pipeline to learn the coefficient of friction of a bar of soap from a real video of the soap sliding on a table. We also learn the coefficient of friction and mass of a Stanford bunny through interactions with a Panda robot arm from synthetic data, and we optimize trajectories in simulation for the Panda arm to push the bunny to a goal location.

Qianzhong Chen, Naixiang Gao, Suning Huang et al. · stanford

Autonomous drones capable of interpreting and executing high-level language instructions in unstructured environments remain a long-standing goal. Yet existing approaches are constrained by their dependence on hand-crafted skills, extensive parameter tuning, or computationally intensive models unsuitable for onboard use. We introduce GRaD-Nav++, a lightweight Vision-Language-Action (VLA) framework that runs fully onboard and follows natural-language commands in real time. Our policy is trained in a photorealistic 3D Gaussian Splatting (3DGS) simulator via Differentiable Reinforcement Learning (DiffRL), enabling efficient learning of low-level control from visual and linguistic inputs. At its core is a Mixture-of-Experts (MoE) action head, which adaptively routes computation to improve generalization while mitigating forgetting. In multi-task generalization experiments, GRaD-Nav++ achieves a success rate of 83% on trained tasks and 75% on unseen tasks in simulation. When deployed on real hardware, it attains 67% success on trained tasks and 50% on unseen ones. In multi-environment adaptation experiments, GRaD-Nav++ achieves an average success rate of 81% across diverse simulated environments and 67% across varied real-world settings. These results establish a new benchmark for fully onboard Vision-Language-Action (VLA) flight and demonstrate that compact, efficient models can enable reliable, language-guided navigation without relying on external infrastructure.

Qianzhong Chen, Justin Yu, Mac Schwager et al. · stanford

Large-scale robot learning has made progress on complex manipulation tasks, yet long horizon, contact rich problems, especially those involving deformable objects, remain challenging due to inconsistent demonstration quality. We propose a stage-aware, video-based reward modeling framework that jointly predicts task stage and fine-grained progress, using natural language subtask annotations to derive consistent labels across variable-length demonstrations. This avoids the brittleness of frame index based labeling and provides stable supervision even in tasks like T-shirt folding. Our reward model is robust to demonstration variability, generalizes to out-of-distribution scenarios, and improves downstream policy training. Building on it, we introduce Reward-Aligned Behavior Cloning (RA-BC), which filters and reweights demonstrations based on reward estimates. Experiments show that our method significantly outperforms baselines in both real-world rollouts and human validation. On T-shirt folding, we achieve 83% success from the flattened state and 67% from the crumpled state, compared to 8% and 0% with vanilla BC. Overall, our results highlight reward modeling as a scalable and annotation-efficient solution for long horizon robotic manipulation. Project website: https://qianzhong-chen.github.io/sarm.github.io/

Jiankai Sun, Yan Xu, Mingyu Ding et al.

Neural Radiance Fields (NeRFs) have become a widely-applied scene representation technique in recent years, showing advantages for robot navigation and manipulation tasks. To further advance the utility of NeRFs for robotics, we propose a transformer-based framework, NeRF-Loc, to extract 3D bounding boxes of objects in NeRF scenes. NeRF-Loc takes a pre-trained NeRF model and camera view as input and produces labeled, oriented 3D bounding boxes of objects as output. Using current NeRF training tools, a robot can train a NeRF environment model in real-time and, using our algorithm, identify 3D bounding boxes of objects of interest within the NeRF for downstream navigation or manipulation tasks. Concretely, we design a pair of paralleled transformer encoder branches, namely the coarse stream and the fine stream, to encode both the context and details of target objects. The encoded features are then fused together with attention layers to alleviate ambiguities for accurate object localization. We have compared our method with conventional RGB(-D) based methods that take rendered RGB images and depths from NeRFs as inputs. Our method is better than the baselines.

Hojune Kim, Timothy Chen, Jiankai Sun et al.

Training vision-language-action (VLA) policies for humanoid loco-manipulation is constrained by the high cost and complexity of collecting human teleoperation demonstrations. VLA policies fine-tuned in simulators have, until now, failed to transfer effectively in humanoid loco-manipulation tasks. We present LEGS (Loco-manipulation via Embodied Gaussian Splatting), a hybrid simulator that composites a mesh foreground (robot, objects, props) over a photorealistic 3D Gaussian Splatting (3DGS) background reconstructed from a handheld scene capture. LEGS uses a procedural motion-primitive generator to synthesize labeled demonstrations at scale without human teleoperation, and a deterministic two-stage color calibration to align the rendered 3DGS image to the robot's deployment camera. On a Unitree G1 humanoid robot, across three pick-and-place tasks of increasing whole-body difficulty and three VLA backbones (psi_0, pi_0.5, GR00T N1.6), a policy trained purely on LEGS data matches or exceeds one trained on human teleoperation demos on every experiment. It also outperforms a mesh-only simulation baseline that ablates the effect of the 3DGS background, showing that photorealistic rendering is a key enabler for synthetic data transfer. Humanoid motion is recorded independently of scene appearance in LEGS, allowing the same auto-generated demonstrations to be re-rendered under new backgrounds and object meshes--covering a new scene at more than 15x lower cost than teleoperation--to augment training data for robustness to scene variations. Under combined object-and-scene appearance shift, the policy trained on re-rendered LEGS-AUG data maintains task success while the baseline trained on teleoperation data fails entirely. Our project page is located at https://legsvla.github.io/.

Kyle Brown, Dylan M. Asmar, Mac Schwager et al.

Mobile autonomous robots have the potential to revolutionize manufacturing processes. However, employing large robot fleets in manufacturing requires addressing challenges including collision-free movement in a shared workspace, effective multi-robot collaboration to manipulate and transport large payloads, complex task allocation due to coupled manufacturing processes, and spatial planning for parallel assembly and transportation of nested subassemblies. We propose a full algorithmic stack for large-scale multi-robot assembly planning that addresses these challenges and can synthesize construction plans for complex assemblies with thousands of parts in a matter of minutes. Our approach takes in a CAD-like product specification and automatically plans a full-stack assembly procedure for a group of robots to manufacture the product. We propose an algorithmic stack that comprises: (i) an iterative radial layout optimization procedure to define a global staging layout for the manufacturing facility, (ii) a graph-repair mixed-integer program formulation and a modified greedy task allocation algorithm to optimally allocate robots and robot sub-teams to assembly and transport tasks, (iii) a geometric heuristic and a hill-climbing algorithm to plan collaborative carrying configurations of robot sub-teams, and (iv) a distributed control policy that enables robots to execute the assembly motion plan collision-free. We also present an open-source multi-robot manufacturing simulator implemented in Julia as a resource to the research community, to test our algorithms and to facilitate multi-robot manufacturing research more broadly. Our empirical results demonstrate the scalability and effectiveness of our approach by generating plans to manufacture a LEGO model of a Saturn V launch vehicle with 1845 parts, 306 subassemblies, and 250 robots in under three minutes on a standard laptop computer.

Joseph A. Vincent, Mac Schwager

Neural networks are increasingly used in robotics as policies, state transition models, state estimation models, or all of the above. With these components being learned from data, it is important to be able to analyze what behaviors were learned and how this affects closed-loop performance. In this paper we take steps toward this goal by developing methods for computing control invariant sets and regions of attraction (ROAs) of dynamical systems represented as neural networks. We focus our attention on feedforward neural networks with the rectified linear unit (ReLU) activation, which are known to implement continuous piecewise-affine (PWA) functions. We describe the Reachable Polyhedral Marching (RPM) algorithm for enumerating the affine pieces of a neural network through an incremental connected walk. We then use this algorithm to compute exact forward and backward reachable sets, from which we provide methods for computing control invariant sets and ROAs. Our approach is unique in that we find these sets incrementally, without Lyapunov-based tools. In our examples we demonstrate the ability of our approach to find non-convex control invariant sets and ROAs on tasks with learned van der Pol oscillator and pendulum models. Further, we provide an accelerated algorithm for computing ROAs that leverages the incremental and connected enumeration of affine regions that RPM provides. We show this acceleration to lead to a 15x speedup in our examples. Finally, we apply our methods to find a set of states that are stabilized by an image-based controller for an aircraft runway control problem.

Jiankai Sun, Shreyas Kousik, David Fridovich-Keil et al. · gatech

Connected autonomous vehicles (CAVs) promise to enhance safety, efficiency, and sustainability in urban transportation. However, this is contingent upon a CAV correctly predicting the motion of surrounding agents and planning its own motion safely. Doing so is challenging in complex urban environments due to frequent occlusions and interactions among many agents. One solution is to leverage smart infrastructure to augment a CAV's situational awareness; the present work leverages a recently proposed "Self-Supervised Traffic Advisor" (SSTA) framework of smart sensors that teach themselves to generate and broadcast useful video predictions of road users. In this work, SSTA predictions are modified to predict future occupancy instead of raw video, which reduces the data footprint of broadcast predictions. The resulting predictions are used within a planning framework, demonstrating that this design can effectively aid CAV motion planning. A variety of numerical experiments study the key factors that make SSTA outputs useful for practical CAV planning in crowded urban environments.

Aiden Swann, Lachlain McGranahan, Hugo Buurmeijer et al.

Vision-Language-Action (VLA) models have emerged as a promising approach for general-purpose robot manipulation. However, their generalization is inconsistent: while these models can perform impressively in some settings, fine-tuned variants often fail on novel objects, scenes, and instructions. We apply mechanistic interpretability techniques to better understand the inner workings of VLA models. To probe internal representations, we train Sparse Autoencoders (SAEs) on hidden layer activations of the VLA. SAEs learn a sparse dictionary whose features act as a compact, interpretable basis for the model's computation. We find that the large majority of extracted SAE features correspond to memorized sequences from specific training demonstrations. However, some features correspond to interpretable, general, and steerable motion primitives and semantic properties, offering a promising glimpse toward VLA generalizability. We propose a metric to categorize features according to whether they represent generalizable transferable primitives or episode-specific memorization. We validate these findings through steering experiments on the LIBERO benchmark. We show that individual SAE features causally influence robot behavior. Steering general features induces behaviors consistent with their semantic meaning and can be applied across tasks and scenes. This work provides the first mechanistic evidence that VLAs can learn generalizable features across tasks and scenes. We observe that supervised fine-tuning on small robotics datasets disproportionately amplifies memorization. In contrast, training on larger, more diverse datasets (e.g., DROID) or using knowledge insulation promotes more general features. We provide an open-source codebase and user-friendly interface for activation collection, SAE training, and feature steering. Our project page is located at http://drvla.github.io

Johnathan Tucker, Denis Liu, Aiden Swann et al.

Vision-Language-Action (VLA) models such as $π_0$ have demonstrated remarkable generalization across diverse fixed-base manipulators. However, transferring these foundation models to aerial platforms remains an open challenge due to the fundamental mismatch between the quasi-static dynamics of fixed-base arms and the underactuated, highly dynamic nature of flight. In this work, we introduce AirVLA, a system that investigates the transferability of manipulation-pretrained VLAs to aerial pick-and-place tasks. We find that while visual representations transfer effectively, the specific control dynamics required for flight do not. To bridge this "dynamics gap" without retraining the foundation model, we introduce a Payload-Aware Guidance mechanism that injects payload constraints directly into the policy's flow-matching sampling process. To overcome data scarcity, we further utilize a Gaussian Splatting pipeline to synthesize navigation training data. We evaluate our method through a cumulative 460 real-world experiments which demonstrate that this synthetic data is a key enabler of performance, unlocking 100% success in navigation tasks where directly fine-tuning on teleoperation data alone attains 81% success. Our inference-time intervention, Payload-Aware Guidance, increases real-world pick-and-place task success from 23% to 50%. Finally, we evaluate the model on a long-horizon compositional task, achieving a 62% overall success rate. These results suggest that pre-trained manipulation VLAs, with appropriate data augmentation and physics-informed guidance, can transfer to aerial manipulation and navigation, as well as the composition of these tasks.

Carlos Plou, Pablo Pueyo, Ruben Martinez-Cantin et al.

Gen-Swarms is an innovative method that leverages and combines the capabilities of deep generative models with reactive navigation algorithms to automate the creation of drone shows. Advancements in deep generative models, particularly diffusion models, have demonstrated remarkable effectiveness in generating high-quality 2D images. Building on this success, various works have extended diffusion models to 3D point cloud generation. In contrast, alternative generative models such as flow matching have been proposed, offering a simple and intuitive transition from noise to meaningful outputs. However, the application of flow matching models to 3D point cloud generation remains largely unexplored. Gen-Swarms adapts these models to automatically generate drone shows. Existing 3D point cloud generative models create point trajectories which are impractical for drone swarms. In contrast, our method not only generates accurate 3D shapes but also guides the swarm motion, producing smooth trajectories and accounting for potential collisions through a reactive navigation algorithm incorporated into the sampling process. For example, when given a text category like Airplane, Gen-Swarms can rapidly and continuously generate numerous variations of 3D airplane shapes. Our experiments demonstrate that this approach is particularly well-suited for drone shows, providing feasible trajectories, creating representative final shapes, and significantly enhancing the overall performance of drone show generation.

Jiankai Sun, Jianing Qiu, Chuanyang Zheng et al.

We seek to accelerate research in developing rich, multimodal scene models trained from egocentric data, based on differentiable volumetric ray-tracing inspired by Neural Radiance Fields (NeRFs). The construction of a NeRF-like model from an egocentric image sequence plays a pivotal role in understanding human behavior and holds diverse applications within the realms of VR/AR. Such egocentric NeRF-like models may be used as realistic simulations, contributing significantly to the advancement of intelligent agents capable of executing tasks in the real-world. The future of egocentric view synthesis may lead to novel environment representations going beyond today's NeRFs by augmenting visual data with multimodal sensors such as IMU for egomotion tracking, audio sensors to capture surface texture and human language context, and eye-gaze trackers to infer human attention patterns in the scene. To support and facilitate the development and evaluation of egocentric multimodal scene modeling, we present a comprehensive multimodal egocentric video dataset. This dataset offers a comprehensive collection of sensory data, featuring RGB images, eye-tracking camera footage, audio recordings from a microphone, atmospheric pressure readings from a barometer, positional coordinates from GPS, connectivity details from Wi-Fi and Bluetooth, and information from dual-frequency IMU datasets (1kHz and 800Hz) paired with a magnetometer. The dataset was collected with the Meta Aria Glasses wearable device platform. The diverse data modalities and the real-world context captured within this dataset serve as a robust foundation for furthering our understanding of human behavior and enabling more immersive and intelligent experiences in the realms of VR, AR, and robotics.

Chan Kim, Keonwoo Kim, Mintaek Oh et al.

Large language models (LLMs) have shown significant potential in guiding embodied agents to execute language instructions across a range of tasks, including robotic manipulation and navigation. However, existing methods are primarily designed for static environments and do not leverage the agent's own experiences to refine its initial plans. Given that real-world environments are inherently stochastic, initial plans based solely on LLMs' general knowledge may fail to achieve their objectives, unlike in static scenarios. To address this limitation, this study introduces the Experience-and-Emotion Map (E2Map), which integrates not only LLM knowledge but also the agent's real-world experiences, drawing inspiration from human emotional responses. The proposed methodology enables one-shot behavior adjustments by updating the E2Map based on the agent's experiences. Our evaluation in stochastic navigation environments, including both simulations and real-world scenarios, demonstrates that the proposed method significantly enhances performance in stochastic environments compared to existing LLM-based approaches. Code and supplementary materials are available at https://e2map.github.io/.

Joseph A. Vincent, Haruki Nishimura, Masha Itkina et al.

With the rise of stochastic generative models in robot policy learning, end-to-end visuomotor policies are increasingly successful at solving complex tasks by learning from human demonstrations. Nevertheless, since real-world evaluation costs afford users only a small number of policy rollouts, it remains a challenge to accurately gauge the performance of such policies. This is exacerbated by distribution shifts causing unpredictable changes in performance during deployment. To rigorously evaluate behavior cloning policies, we present a framework that provides a tight lower-bound on robot performance in an arbitrary environment, using a minimal number of experimental policy rollouts. Notably, by applying the standard stochastic ordering to robot performance distributions, we provide a worst-case bound on the entire distribution of performance (via bounds on the cumulative distribution function) for a given task. We build upon established statistical results to ensure that the bounds hold with a user-specified confidence level and tightness, and are constructed from as few policy rollouts as possible. In experiments we evaluate policies for visuomotor manipulation in both simulation and hardware. Specifically, we (i) empirically validate the guarantees of the bounds in simulated manipulation settings, (ii) find the degree to which a learned policy deployed on hardware generalizes to new real-world environments, and (iii) rigorously compare two policies tested in out-of-distribution settings. Our experimental data, code, and implementation of confidence bounds are open-source.

Chuanyang Zheng, Jiankai Sun, Yihang Gao et al.

The placement of normalization layers, specifically Pre-Norm and Post-Norm, remains an open question in Transformer architecture design. In this work, we rethink these approaches through the lens of manifold optimization, interpreting the outputs of the Feed-Forward Network (FFN) and attention layers as update directions in optimization. Building on this perspective, we introduce GeoNorm, a novel method that replaces standard normalization with geodesic updates on the manifold. Furthermore, analogous to learning rate schedules, we propose a layer-wise update decay for the FFN and attention components. Comprehensive experiments demonstrate that GeoNorm consistently outperforms existing normalization methods in Transformer models. Crucially, GeoNorm can be seamlessly integrated into standard Transformer architectures, achieving performance improvements with negligible additional computational cost.

Timothy Chen, Adam Dai, Maximilian Adang et al.

What makes a good viewpoint? The quality of the data used to learn 3D reconstructions is crucial for enabling efficient and accurate scene modeling. We study the active view selection problem and develop a principled analysis that yields a simple and interpretable criterion for selecting informative camera poses. Our key insight is that informative views can be obtained by minimizing a tractable approximation of the Fisher Information Gain, which reduces to favoring viewpoints that cover geometry that has been insufficiently observed by past cameras. This leads to a lightweight coverage-based view selection metric that avoids expensive transmittance estimation and is robust to noise and training dynamics. We call this metric COVER (Camera Optimization for View Exploration and Reconstruction). We integrate our method into the Nerfstudio framework and evaluate it on real datasets within fixed and embodied data acquisition scenarios. Across multiple datasets and radiance-field baselines, our method consistently improves reconstruction quality compared to state-of-the-art active view selection methods. Additional visualizations and our Nerfstudio package can be found at https://chengine.github.io/nbv_gym/.

Kunal Shah, Guillermo Angeris, Mac Schwager

We present a general decentralized formulation for a large class of collision avoidance methods and show that all collision avoidance methods of this form are guaranteed to be collision free. This class includes several existing algorithms in the literature as special cases. We then present a particular instance of this collision avoidance method, CARP (Collision Avoidance by Reciprocal Projections), that is effective even when the estimates of other agents' positions and velocities are noisy. The method's main computational step involves the solution of a small convex optimization problem, which can be quickly solved in practice, even on embedded platforms, making it practical to use on computationally-constrained robots such as quadrotors. This method can be extended to find smooth polynomial trajectories for higher dynamic systems such at quadrotors. We demonstrate this algorithm's performance in simulations and on a team of physical quadrotors. Our method finds optimal projections in a median time of 17.12ms for 285 instances of 100 randomly generated obstacles, and produces safe polynomial trajectories at over 60hz on-board quadrotors. Our paper is accompanied by an open source Julia implementation and ROS package.

Joseph A. Vincent, Mac Schwager

We present a method for computing exact reachable sets for deep neural networks with rectified linear unit (ReLU) activation. Our method is well-suited for use in rigorous safety analysis of robotic perception and control systems with deep neural network components. Our algorithm can compute both forward and backward reachable sets for a ReLU network iterated over multiple time steps, as would be found in a perception-action loop in a robotic system. Our algorithm is unique in that it builds the reachable sets by incrementally enumerating polyhedral cells in the input space, rather than iterating layer-by-layer through the network as in other methods. If an unsafe cell is found, our algorithm can return this result without completing the full reachability computation, thus giving an anytime property that accelerates safety verification. In addition, our method requires less memory during execution compared to existing methods where memory can be a limiting factor. We demonstrate our algorithm on safety verification of the ACAS Xu aircraft advisory system. We find unsafe actions many times faster than the fastest existing method and certify no unsafe actions exist in about twice the time of the existing method. We also compute forward and backward reachable sets for a learned model of pendulum dynamics over a 50 time step horizon in 87s on a laptop computer. Algorithm source code: https://github.com/StanfordMSL/Neural-Network-Reach.

Simon Le Cleac'h, Mac Schwager, Zachary Manchester

Existing game-theoretic planning methods assume that the robot knows the objective functions of the other agents a priori while, in practical scenarios, this is rarely the case. This paper introduces LUCIDGames, an inverse optimal control algorithm that is able to estimate the other agents' objective functions in real time, and incorporate those estimates online into a receding-horizon game-theoretic planner. LUCIDGames solves the inverse optimal control problem by recasting it in a recursive parameter-estimation framework. LUCIDGames uses an unscented Kalman filter (UKF) to iteratively update a Bayesian estimate of the other agents' cost function parameters, improving that estimate online as more data is gathered from the other agents' observed trajectories. The planner then takes account of the uncertainty in the Bayesian parameter estimates of other agents by planning a trajectory for the robot subject to uncertainty ellipse constraints. The algorithm assumes no explicit communication or coordination between the robot and the other agents in the environment. An MPC implementation of LUCIDGames demonstrates real-time performance on complex autonomous driving scenarios with an update frequency of 40 Hz. Empirical results demonstrate that LUCIDGames improves the robot's performance over existing game-theoretic and traditional MPC planning approaches. Our implementation of LUCIDGames is available at https://github.com/RoboticExplorationLab/LUCIDGames.jl.

Pablo Pueyo, Eric Cristofalo, Eduardo Montijano et al.

Drones and Unmanned Aerial Vehicles (UAV's) are becoming increasingly popular in the film and entertainment industries in part because of their maneuverability and the dynamic shots and perspectives they enable. While there exists methods for controlling the position and orientation of the drones for visibility, other artistic elements of the filming process, such as focal blur and light control, remain unexplored in the robotics community. The lack of cinemetographic robotics solutions is partly due to the cost associated with the cameras and devices used in the filming industry, but also because state-of-the-art photo-realistic robotics simulators only utilize a full in-focus pinhole camera model which does incorporate these desired artistic attributes. To overcome this, the main contribution of this work is to endow the well-known drone simulator, AirSim, with a cinematic camera as well as extended its API to control all of its parameters in real time, including various filming lenses and common cinematographic properties. In this paper, we detail the implementation of our AirSim modification, CinemAirSim, present examples that illustrate the potential of the new tool, and highlight the new research opportunities that the use of cinematic cameras can bring to research in robotics and control. https://github.com/ppueyor/CinematicAirSim

Chuanyang Zheng, Jiankai Sun, Yihang Gao et al.

Since its introduction in 2017, the Transformer has become one of the most widely adopted architectures in modern deep learning. Despite extensive efforts to improve positional encoding, attention mechanisms, and feed-forward networks, the core token-mixing mechanism in Transformers remains attention. In this work, we show that the attention module in Transformers can be interpreted as performing Nadaraya-Watson regression, where it computes similarities between tokens and aggregates the corresponding values accordingly. Motivated by this perspective, we propose Cubit, a potential next-generation architecture that leverages Kernel Ridge Regression (KRR), while the vanilla Transformer relies on Nadaraya-Watson regression. Specifically, Cubit modifies the classical attention computation by incorporating the closed-form solution of KRR, combining value aggregation through kernel similarities with normalization via the inverse of the kernel matrix. To improve the training stability, we further propose the Limited-Range Rescale (LRR), which rescales the value layer within a controlled range. We argue that Cubit, as a KRR-based architecture, provides a stronger mathematical foundation than the vanilla Transformer, whose attention mechanism corresponds to Nadaraya-Watson regression. We validate this claim through comprehensive experiments. The experimental results suggest that Cubit may exhibit stronger long-sequence modeling capability. In particular, its performance gain over the Transformer appears to increase as the training sequence length grows.

Ola Shorinwa, Johnathan Tucker, Aliyah Smith et al.

We present Splat-MOVER, a modular robotics stack for open-vocabulary robotic manipulation, which leverages the editability of Gaussian Splatting (GSplat) scene representations to enable multi-stage manipulation tasks. Splat-MOVER consists of: (i) ASK-Splat, a GSplat representation that distills semantic and grasp affordance features into the 3D scene. ASK-Splat enables geometric, semantic, and affordance understanding of 3D scenes, which is critical in many robotics tasks; (ii) SEE-Splat, a real-time scene-editing module using 3D semantic masking and infilling to visualize the motions of objects that result from robot interactions in the real-world. SEE-Splat creates a "digital twin" of the evolving environment throughout the manipulation task; and (iii) Grasp-Splat, a grasp generation module that uses ASK-Splat and SEE-Splat to propose affordance-aligned candidate grasps for open-world objects. ASK-Splat is trained in real-time from RGB images in a brief scanning phase prior to operation, while SEE-Splat and Grasp-Splat run in real-time during operation. We demonstrate the superior performance of Splat-MOVER in hardware experiments on a Kinova robot compared to two recent baselines in four single-stage, open-vocabulary manipulation tasks and in four multi-stage manipulation tasks, using the edited scene to reflect changes due to prior manipulation stages, which is not possible with existing baselines. Video demonstrations and the code for the project are available at https://splatmover.github.io.

JunEn Low, Maximilian Adang, Javier Yu et al.

We propose a new simulator, training approach, and policy architecture, collectively called SOUS VIDE, for end-to-end visual drone navigation. Our trained policies exhibit zero-shot sim-to-real transfer with robust real-world performance using only onboard perception and computation. Our simulator, called FiGS, couples a computationally simple drone dynamics model with a high visual fidelity Gaussian Splatting scene reconstruction. FiGS can quickly simulate drone flights producing photorealistic images at up to 130 fps. We use FiGS to collect 100k-300k image/state-action pairs from an expert MPC with privileged state and dynamics information, randomized over dynamics parameters and spatial disturbances. We then distill this expert MPC into an end-to-end visuomotor policy with a lightweight neural architecture, called SV-Net. SV-Net processes color image, optical flow and IMU data streams into low-level thrust and body rate commands at 20 Hz onboard a drone. Crucially, SV-Net includes a learned module for low-level control that adapts at runtime to variations in drone dynamics. In a campaign of 105 hardware experiments, we show SOUS VIDE policies to be robust to 30% mass variations, 40 m/s wind gusts, 60% changes in ambient brightness, shifting or removing objects from the scene, and people moving aggressively through the drone's visual field. Code, data, and experiment videos can be found on our project page: https://stanfordmsl.github.io/SousVide/.

Ola Shorinwa, Jiankai Sun, Mac Schwager et al.

We present SIREN for registration of multi-robot Gaussian Splatting (GSplat) maps, with zero access to camera poses, images, and inter-map transforms for initialization or fusion of local submaps. To realize these capabilities, SIREN harnesses the versatility and robustness of semantics in three critical ways to derive a rigorous registration pipeline for multi-robot GSplat maps. First, SIREN utilizes semantics to identify feature-rich regions of the local maps where the registration problem is better posed, eliminating the need for any initialization which is generally required in prior work. Second, SIREN identifies candidate correspondences between Gaussians in the local maps using robust semantic features, constituting the foundation for robust geometric optimization, coarsely aligning 3D Gaussian primitives extracted from the local maps. Third, this key step enables subsequent photometric refinement of the transformation between the submaps, where SIREN leverages novel-view synthesis in GSplat maps along with a semantics-based image filter to compute a high-accuracy non-rigid transformation for the generation of a high-fidelity fused map. We demonstrate the superior performance of SIREN compared to competing baselines across a range of real-world datasets, and in particular, across the most widely-used robot hardware platforms, including a manipulator, drone, and quadruped. In our experiments, SIREN achieves about 90x smaller rotation errors, 300x smaller translation errors, and 44x smaller scale errors in the most challenging scenes, where competing methods struggle. We will release the code and provide a link to the project page after the review process.

Ola Shorinwa, Jiankai Sun, Mac Schwager

We present FAST-Splat for fast, ambiguity-free semantic Gaussian Splatting, which seeks to address the main limitations of existing semantic Gaussian Splatting methods, namely: slow training and rendering speeds; high memory usage; and ambiguous semantic object localization. We take a bottom-up approach in deriving FAST-Splat, dismantling the limitations of closed-set semantic distillation to enable open-set (open-vocabulary) semantic distillation. Ultimately, this key approach enables FAST-Splat to provide precise semantic object localization results, even when prompted with ambiguous user-provided natural-language queries. Further, by exploiting the explicit form of the Gaussian Splatting scene representation to the fullest extent, FAST-Splat retains the remarkable training and rendering speeds of Gaussian Splatting. Precisely, while existing semantic Gaussian Splatting methods distill semantics into a separate neural field or utilize neural models for dimensionality reduction, FAST-Splat directly augments each Gaussian with specific semantic codes, preserving the training, rendering, and memory-usage advantages of Gaussian Splatting over neural field methods. These Gaussian-specific semantic codes, together with a hash-table, enable semantic similarity to be measured with open-vocabulary user prompts and further enable FAST-Splat to respond with unambiguous semantic object labels and $3$D masks, unlike prior methods. In experiments, we demonstrate that FAST-Splat is 6x to 8x faster to train, achieves between 18x to 51x faster rendering speeds, and requires about 6x smaller GPU memory, compared to the best-competing semantic Gaussian Splatting methods. Further, FAST-Splat achieves relatively similar or better semantic segmentation performance compared to existing methods. After the review period, we will provide links to the project website and the codebase.

Suning Huang, Jiaqi Shao, Ke Wang et al.

Have you ever post-trained a generalist vision-language-action (VLA) policy on a small demonstration dataset, only to find that it stops responding to new instructions and is limited to behaviors observed during post-training? We identify this phenomenon as lock-in: after low-data, supervised fine-tuning (SFT), the policy becomes overly specialized to the post-training data and fails to generalize to novel instructions, manifesting as concept lock-in (fixation on training objects/attributes) and spatial lock-in (fixation on training spatial targets). Many existing remedies introduce additional supervision signals, such as those derived from foundation models or auxiliary objectives, or rely on augmented datasets to recover generalization. In this paper, we show that the policy's internal pre-trained knowledge is sufficient: DeLock mitigates lock-in by preserving visual grounding during post-training and applying test-time contrastive prompt guidance to steer the policy's denoising dynamics according to novel instructions. Across eight simulation and real-world evaluations, DeLock consistently outperforms strong baselines and matches or exceeds the performance of a state-of-the-art generalist policy post-trained with substantially more curated demonstrations.

Pablo Pueyo, Eduardo Montijano, Ana C. Murillo et al.

This paper introduces CLIPSwarm, a new algorithm designed to automate the modeling of swarm drone formations based on natural language. The algorithm begins by enriching a provided word, to compose a text prompt that serves as input to an iterative approach to find the formation that best matches the provided word. The algorithm iteratively refines formations of robots to align with the textual description, employing different steps for "exploration" and "exploitation". Our framework is currently evaluated on simple formation targets, limited to contour shapes. A formation is visually represented through alpha-shape contours and the most representative color is automatically found for the input word. To measure the similarity between the description and the visual representation of the formation, we use CLIP [1], encoding text and images into vectors and assessing their similarity. Subsequently, the algorithm rearranges the formation to visually represent the word more effectively, within the given constraints of available drones. Control actions are then assigned to the drones, ensuring robotic behavior and collision-free movement. Experimental results demonstrate the system's efficacy in accurately modeling robot formations from natural language descriptions. The algorithm's versatility is showcased through the execution of drone shows in photorealistic simulation with varying shapes. We refer the reader to the supplementary video for a visual reference of the results.

Ruiyang Hao, Haibao Yu, Jiaru Zhong et al.

With the rapid advancement of autonomous driving technology, vehicle-to-everything (V2X) communication has emerged as a key enabler for extending perception range and enhancing driving safety by providing visibility beyond the line of sight. However, integrating multi-source sensor data from both ego-vehicles and infrastructure under real-world constraints, such as limited communication bandwidth and dynamic environments, presents significant technical challenges. To facilitate research in this area, we organized the End-to-End Autonomous Driving through V2X Cooperation Challenge, which features two tracks: cooperative temporal perception and cooperative end-to-end planning. Built on the UniV2X framework and the V2X-Seq-SPD dataset, the challenge attracted participation from over 30 teams worldwide and established a unified benchmark for evaluating cooperative driving systems. This paper describes the design and outcomes of the challenge, highlights key research problems including bandwidth-aware fusion, robust multi-agent planning, and heterogeneous sensor integration, and analyzes emerging technical trends among top-performing solutions. By addressing practical constraints in communication and data fusion, the challenge contributes to the development of scalable and reliable V2X-cooperative autonomous driving systems.

Chuanyang Zheng, Jiankai Sun, Yihang Gao et al.

Mixture-of-Experts (MoE) has become a cornerstone in recent state-of-the-art large language models (LLMs). Traditionally, MoE relies on $\mathrm{Softmax}$ as the router score function to aggregate expert output, a designed choice that has persisted from the earliest MoE models to modern LLMs, and is now widely regarded as standard practice. However, the necessity of using $\mathrm{Softmax}$ to project router weights into a probability simplex remains an unchallenged assumption rather than a principled design choice. In this work, we first revisit the classical Nadaraya-Watson regression and observe that MoE shares the same mathematical formulation as Nadaraya-Watson regression. Furthermore, we show that both feed-forward neural network (FFN) and MoE can be interpreted as a special case of Nadaraya-Watson regression, where the kernel function corresponds to the input neurons of the output layer. Motivated by these insights, we propose the \textbf{zero-additional-cost} Kernel Inspired Router with Normalization (KERN), an FFN-style router function, as an alternative to $\mathrm{Softmax}$. We demonstrate that this router generalizes both $\mathrm{Sigmoid}$- and $\mathrm{Softmax}$-based routers. \textbf{Based on empirical observations and established practices in FFN implementation, we recommend the use of $\mathrm{ReLU}$ activation and $\ell_2$-normalization in $\mathrm{KERN}$ router function.} Comprehensive experiments in MoE and LLM validate the effectiveness of the proposed FFN-style router function \methodNorm.

Aishwarya Mandyam, Jason Meng, Ge Gao et al.

Off-policy evaluation (OPE) methods aim to estimate the value of a new reinforcement learning (RL) policy prior to deployment. Recent advances have shown that leveraging auxiliary datasets, such as those synthesized by generative models, can improve the accuracy of these value estimates. Unfortunately, such auxiliary datasets may also be biased, and existing methods for using data augmentation for OPE in RL lack principled uncertainty quantification. In high stakes settings like healthcare, reliable uncertainty estimates are important for comparing policy value estimates. In this work, we propose two approaches to construct valid confidence intervals for OPE when using data augmentation. The first provides a confidence interval over the policy performance conditioned on a particular initial state $V^π(s_0)$-- such intervals are particularly important for human-centered applications. To do so we introduce a new conformal prediction method for high dimensional state MDPs. Second, we consider the more common task of estimating the average policy performance over many initial states; to do so we draw on ideas from doubly robust estimation and prediction powered inference. Across simulators spanning robotics, healthcare and inventory management, and a real healthcare dataset from MIMIC-IV, we find that our methods can use augmented data and still consistently produce intervals that cover the ground truth values, unlike previously proposed methods.

Isaac R. Ward, Michelle Ho, Houjun Liu et al.

Deploying visuomotor robots at scale is challenging due to the potential for anomalous failures to degrade performance, cause damage, or endanger human life. Bimanual manipulators are no exception; these robots have vast state spaces comprised of high-dimensional images and proprioceptive signals. Explicitly defining failure modes within such state spaces is infeasible. In this work, we overcome these challenges by training a probabilistic, history informed, world model within the compressed latent space of a pretrained vision foundation model (NVIDIA's Cosmos Tokenizer). The model outputs uncertainty estimates alongside its predictions that serve as non-conformity scores within a conformal prediction framework. We use these scores to develop a runtime monitor, correlating periods of high uncertainty with anomalous failures. To test these methods, we use the simulated Push-T environment and the Bimanual Cable Manipulation dataset, the latter of which we introduce in this work. This new dataset features trajectories with multiple synchronized camera views, proprioceptive signals, and annotated failures from a challenging data center maintenance task. We benchmark our methods against baselines from the anomaly detection and out-of-distribution detection literature, and show that our approach considerably outperforms statistical techniques. Furthermore, we show that our approach requires approximately one twentieth of the trainable parameters as the next-best learning-based approach, yet outperforms it by 3.8% in terms of failure detection rate, paving the way toward safely deploying manipulator robots in real-world environments where reliability is non-negotiable.

Aaron O. Feldman, D. Isaiah Harp, Joseph Duncan et al.

We develop a data-driven approach for runtime safety monitoring in flight testing, where pilots perform maneuvers on aircraft with uncertain parameters. Because safety violations can arise unexpectedly as a result of these uncertainties, pilots need clear, preemptive criteria to abort the maneuver in advance of safety violation. To solve this problem, we use offline stochastic trajectory simulation to learn a calibrated statistical model of the short-term safety risk facing pilots. We use flight testing as a motivating example for data-driven learning/monitoring of safety due to its inherent safety risk, uncertainty, and human-interaction. However, our approach consists of three broadly-applicable components: a model to predict future state from recent observations, a nearest neighbor model to classify the safety of the predicted state, and classifier calibration via conformal prediction. We evaluate our method on a flight dynamics model with uncertain parameters, demonstrating its ability to reliably identify unsafe scenarios, match theoretical guarantees, and outperform baseline approaches in preemptive classification of risk.

Maggie Wang, Stephen Tian, Aiden Swann et al.

Learning robotic manipulation policies directly in the real world can be expensive and time-consuming. While reinforcement learning (RL) policies trained in simulation present a scalable alternative, effective sim-to-real transfer remains challenging, particularly for tasks that require precise dynamics. To address this, we propose Phys2Real, a real-to-sim-to-real RL pipeline that combines vision-language model (VLM)-inferred physical parameter estimates with interactive adaptation through uncertainty-aware fusion. Our approach consists of three core components: (1) high-fidelity geometric reconstruction with 3D Gaussian splatting, (2) VLM-inferred prior distributions over physical parameters, and (3) online physical parameter estimation from interaction data. Phys2Real conditions policies on interpretable physical parameters, refining VLM predictions with online estimates via ensemble-based uncertainty quantification. On planar pushing tasks of a T-block with varying center of mass (CoM) and a hammer with an off-center mass distribution, Phys2Real achieves substantial improvements over a domain randomization baseline: 100% vs 79% success rate for the bottom-weighted T-block, 57% vs 23% in the challenging top-weighted T-block, and 15% faster average task completion for hammer pushing. Ablation studies indicate that the combination of VLM and interaction information is essential for success. Project website: https://phys2real.github.io/ .

Sixten Norelius, Aaron O. Feldman, Mac Schwager

We propose SketchPlan, a diffusion-based planner that interprets 2D hand-drawn sketches over depth images to generate 3D flight paths for drone navigation. SketchPlan comprises two components: a SketchAdapter that learns to map the human sketches to projected 2D paths, and DiffPath, a diffusion model that infers 3D trajectories from 2D projections and a first person view depth image. Our model achieves zero-shot sim-to-real transfer, generating accurate and safe flight paths in previously unseen real-world environments. To train the model, we build a synthetic dataset of 32k flight paths using a diverse set of photorealistic 3D Gaussian Splatting scenes. We automatically label the data by computing 2D projections of the 3D flight paths onto the camera plane, and use this to train the DiffPath diffusion model. However, since real human 2D sketches differ significantly from ideal 2D projections, we additionally label 872 of the 3D flight paths with real human sketches and use this to train the SketchAdapter to infer the 2D projection from the human sketch. We demonstrate SketchPlan's effectiveness in both simulated and real-world experiments, and show through ablations that training on a mix of human labeled and auto-labeled data together with a modular design significantly boosts its capabilities to correctly interpret human intent and infer 3D paths. In real-world drone tests, SketchPlan achieved 100\% success in low/medium clutter and 40\% in unseen high-clutter environments, outperforming key ablations by 20-60\% in task completion.

Chuanyang Zheng, Jiankai Sun, Yihang Gao et al.

The attention mechanism is a core component of the Transformer architecture. Various methods have been developed to compute attention scores, including multi-head attention (MHA), multi-query attention, group-query attention and so on. We further analyze the MHA and observe that its performance improves as the number of attention heads increases, provided the hidden size per head remains sufficiently large. Therefore, increasing both the head count and hidden size per head with minimal parameter overhead can lead to significant performance gains at a low cost. Motivated by this insight, we introduce Simulated Attention Score (SAS), which maintains a compact model size while simulating a larger number of attention heads and hidden feature dimension per head. This is achieved by projecting a low-dimensional head representation into a higher-dimensional space, effectively increasing attention capacity without increasing parameter count. Beyond the head representations, we further extend the simulation approach to feature dimension of the key and query embeddings, enhancing expressiveness by mimicking the behavior of a larger model while preserving the original model size. To control the parameter cost, we also propose Parameter-Efficient Attention Aggregation (PEAA). Comprehensive experiments on a variety of datasets and tasks demonstrate the effectiveness of the proposed SAS method, achieving significant improvements over different attention variants.

Aiden Swann, Matthew Strong, Won Kyung Do et al.

In this work, we propose a novel method to supervise 3D Gaussian Splatting (3DGS) scenes using optical tactile sensors. Optical tactile sensors have become widespread in their use in robotics for manipulation and object representation; however, raw optical tactile sensor data is unsuitable to directly supervise a 3DGS scene. Our representation leverages a Gaussian Process Implicit Surface to implicitly represent the object, combining many touches into a unified representation with uncertainty. We merge this model with a monocular depth estimation network, which is aligned in a two stage process, coarsely aligning with a depth camera and then finely adjusting to match our touch data. For every training image, our method produces a corresponding fused depth and uncertainty map. Utilizing this additional information, we propose a new loss function, variance weighted depth supervised loss, for training the 3DGS scene model. We leverage the DenseTact optical tactile sensor and RealSense RGB-D camera to show that combining touch and vision in this manner leads to quantitatively and qualitatively better results than vision or touch alone in a few-view scene syntheses on opaque as well as on reflective and transparent objects. Please see our project page at http://armlabstanford.github.io/touch-gs

Negar Mehr, Mingyu Wang, Mac Schwager

In this paper, we study the problem of multiple stochastic agents interacting in a dynamic game scenario with continuous state and action spaces. We define a new notion of stochastic Nash equilibrium for boundedly rational agents, which we call the Entropic Cost Equilibrium (ECE). We show that ECE is a natural extension to multiple agents of Maximum Entropy optimality for single agents. We solve both the "forward" and "inverse" problems for the multi-agent ECE game. For the forward problem, we provide a Riccati algorithm to compute closed-form ECE feedback policies for the agents, which are exact in the Linear-Quadratic-Gaussian case. We give an iterative variant to find locally ECE feedback policies for the nonlinear case. For the inverse problem, we present an algorithm to infer the cost functions of the multiple interacting agents given noisy, boundedly rational input and state trajectory examples from agents acting in an ECE. The effectiveness of our algorithms is demonstrated in a simulated multi-agent collision avoidance scenario, and with data from the INTERACTION traffic dataset. In both cases, we show that, by taking into account the agents' game theoretic interactions using our algorithm, a more accurate model of agents' costs can be learned, compared with standard inverse optimal control methods.

Michal Adamkiewicz, Timothy Chen, Adam Caccavale et al.

Neural Radiance Fields (NeRFs) have recently emerged as a powerful paradigm for the representation of natural, complex 3D scenes. NeRFs represent continuous volumetric density and RGB values in a neural network, and generate photo-realistic images from unseen camera viewpoints through ray tracing. We propose an algorithm for navigating a robot through a 3D environment represented as a NeRF using only an on-board RGB camera for localization. We assume the NeRF for the scene has been pre-trained offline, and the robot's objective is to navigate through unoccupied space in the NeRF to reach a goal pose. We introduce a trajectory optimization algorithm that avoids collisions with high-density regions in the NeRF based on a discrete time version of differential flatness that is amenable to constraining the robot's full pose and control inputs. We also introduce an optimization based filtering method to estimate 6DoF pose and velocities for the robot in the NeRF given only an onboard RGB camera. We combine the trajectory planner with the pose filter in an online replanning loop to give a vision-based robot navigation pipeline. We present simulation results with a quadrotor robot navigating through a jungle gym environment, the inside of a church, and Stonehenge using only an RGB camera. We also demonstrate an omnidirectional ground robot navigating through the church, requiring it to reorient to fit through the narrow gap. Videos of this work can be found at https://mikh3x4.github.io/nerf-navigation/ .

Fengjun Yang, Negar Mehr, Mac Schwager

We propose a method, based on empirical game theory, for a robot operating as part of a team to choose its role within the team without explicitly communicating with team members, by leveraging its knowledge about the team structure. To do this, we formulate the role assignment problem as a dynamic game, and borrow tools from empirical game-theoretic analysis to analyze such games. Based on this game-theoretic formulation, we propose a distributed controller for each robot to dynamically decide on the best role to take. We demonstrate our method in simulations of a collaborative planar manipulation scenario in which each agent chooses from a set of feedback control policies at each instant. The agents can effectively collaborate without communication to manipulate the object while also avoiding collisions using our method.

Claire Chen, Preston Culbertson, Marion Lepert et al.

Dexterous manipulation tasks often require contact switching, where fingers make and break contact with the object. We propose a method that plans trajectories for dexterous manipulation tasks involving contact switching using contact-implicit trajectory optimization (CITO) augmented with a high-level discrete contact sequence planner. We first use the high-level planner to find a sequence of finger contact switches given a desired object trajectory. With this contact sequence plan, we impose additional constraints in the CITO problem. We show that our method finds trajectories approximately 7 times faster than a general CITO baseline for a four-finger planar manipulation scenario. Furthermore, when executing the planned trajectories in a full dynamics simulator, we are able to more closely track the object pose trajectories planned by our method than those planned by the baselines.

Javier Yu, Joseph A. Vincent, Mac Schwager

We present a distributed algorithm that enables a group of robots to collaboratively optimize the parameters of a deep neural network model while communicating over a mesh network. Each robot only has access to its own data and maintains its own version of the neural network, but eventually learns a model that is as good as if it had been trained on all the data centrally. No robot sends raw data over the wireless network, preserving data privacy and ensuring efficient use of wireless bandwidth. At each iteration, each robot approximately optimizes an augmented Lagrangian function, then communicates the resulting weights to its neighbors, updates dual variables, and repeats. Eventually, all robots' local network weights reach a consensus. For convex objective functions, we prove this consensus is a global optimum. We compare our algorithm to two existing distributed deep neural network training algorithms in (i) an MNIST image classification task, (ii) a multi-robot implicit mapping task, and (iii) a multi-robot reinforcement learning task. In all of our experiments our method out performed baselines, and was able to achieve validation loss equivalent to centrally trained models. See \href{https://msl.stanford.edu/projects/dist_nn_train}{https://msl.stanford.edu/projects/dist\_nn\_train} for videos and a link to our GitHub repository.

Nathan S. Usevitch, Trevor Halsted, Zachary M. Hammond et al.

Truss robots, or robots that consist of extensible links connected at universal joints, are often designed with modular physical components but require centralized control techniques. This paper presents a distributed control technique for truss robots. The truss robot is viewed as a collective, where each individual node of the robot is capable of measuring the lengths of the neighboring edges, communicating with a subset of the other nodes, and computing and executing its own control actions with its connected edges. Through an iterative distributed optimization, the individual members utilize local information to converge on a global estimate of the robot's state, and then coordinate their planned motion to achieve desired global behavior. This distributed optimization is based on a consensus alternating direction method of multipliers framework. This distributed algorithm is then adapted to control an isoperimetric truss robot, and the distributed algorithm is used in an experimental demonstration. The demonstration allows a user to broadcast commands to a single node of the robot, which then ensures the coordinated motion of all other nodes to achieve the desired global motion.

Simon Le Cleac'h, Taylor Howell, Shuo Yang et al.

We present a general approach for controlling robotic systems that make and break contact with their environments. Contact-implicit model predictive control (CI-MPC) generalizes linear MPC to contact-rich settings by utilizing a bi-level planning formulation with lower-level contact dynamics formulated as time-varying linear complementarity problems (LCPs) computed using strategic Taylor approximations about a reference trajectory. These dynamics enable the upper-level planning problem to reason about contact timing and forces, and generate entirely new contact-mode sequences online. To achieve reliable and fast numerical convergence, we devise a structure-exploiting interior-point solver for these LCP contact dynamics and a custom trajectory optimizer for the tracking problem. We demonstrate real-time solution rates for CI-MPC and the ability to generate and track non-periodic behaviours in hardware experiments on a quadrupedal robot. We also show that the controller is robust to model mismatch and can respond to disturbances by discovering and exploiting new contact modes across a variety of robotic systems in simulation, including a pushbot, planar hopper, planar quadruped, and planar biped.

Simon Le Cleac'h, Mac Schwager, Zachary Manchester

Dynamic games are an effective paradigm for dealing with the control of multiple interacting actors. This paper introduces ALGAMES (Augmented Lagrangian GAME-theoretic Solver), a solver that handles trajectory-optimization problems with multiple actors and general nonlinear state and input constraints. Its novelty resides in satisfying the first-order optimality conditions with a quasi-Newton root-finding algorithm and rigorously enforcing constraints using an augmented Lagrangian method. We evaluate our solver in the context of autonomous driving on scenarios with a strong level of interactions between the vehicles. We assess the robustness of the solver using Monte Carlo simulations. It is able to reliably solve complex problems like ramp merging with three vehicles three times faster than a state-of-the-art DDP-based approach. A model-predictive control (MPC) implementation of the algorithm, running at more than 60 Hz, demonstrates ALGAMES' ability to mitigate the "frozen robot" problem on complex autonomous driving scenarios like merging onto a crowded highway.

Pablo Pueyo, Eduardo Montijano, Ana C. Murillo et al.

We present CineMPC, an algorithm to autonomously control a UAV-borne video camera in a nonlinear Model Predicted Control (MPC) loop. CineMPC controls both the position and orientation of the camera -- the camera extrinsics -- as well as the lens focal length, focal distance, and aperture -- the camera intrinsics. While some existing solutions autonomously control the position and orientation of the camera, no existing solutions also control the intrinsic parameters, which are essential tools for rich cinematographic expression. The intrinsic parameters control the parts of the scene that are focused or blurred, the viewers' perception of depth in the scene and the position of the targets in the image. CineMPC closes the loop from camera images to UAV trajectory and lens parameters in order to follow the desired relative trajectory and image composition as the targets move through the scene. Experiments using a photo-realistic environment demonstrate the capabilities of the proposed control framework to successfully achieve a full array of cinematographic effects not possible without full camera control.

Trevor Halsted, Ola Shorinwa, Javier Yu et al.

Distributed optimization consists of multiple computation nodes working together to minimize a common objective function through local computation iterations and network-constrained communication steps. In the context of robotics, distributed optimization algorithms can enable multi-robot systems to accomplish tasks in the absence of centralized coordination. We present a general framework for applying distributed optimization as a module in a robotics pipeline. We survey several classes of distributed optimization algorithms and assess their practical suitability for multi-robot applications. We further compare the performance of different classes of algorithms in simulations for three prototypical multi-robot problem scenarios. The Consensus Alternating Direction Method of Multipliers (C-ADMM) emerges as a particularly attractive and versatile distributed optimization method for multi-robot systems.

Claire Chen, Krishnan Srinivasan, Jeffrey Zhang et al.

This report describes our approach for Phase 3 of the Real Robot Challenge. To solve cuboid manipulation tasks of varying difficulty, we decompose each task into the following primitives: moving the fingers to the cuboid to grasp it, turning it on the table to minimize orientation error, and re-positioning it to the goal position. We use model-based trajectory optimization and control to plan and execute these primitives. These grasping, turning, and re-positioning primitives are sequenced with a state-machine that determines which primitive to execute given the current object state and goal. Our method shows robust performance over multiple runs with randomized initial and goal positions. With this approach, our team placed second in the challenge, under the anonymous name "sombertortoise" on the leaderboard. Example runs of our method solving each of the four levels can be seen in this video (https://www.youtube.com/watch?v=I65Kwu9PGmg&list=PLt9QxrtaftrHGXcp4Oh8-s_OnQnBnLtei&index=1).

Benjamin Ramtoula, Adam Caccavale, Giovanni Beltrame et al.

Determining the relative position and orientation of objects in an environment is a fundamental building block for a wide range of robotics applications. To accomplish this task efficiently in practical settings, a method must be fast, use common sensors, and generalize easily to new objects and environments. We present MSL-RAPTOR, a two-stage algorithm for tracking a rigid body with a monocular camera. The image is first processed by an efficient neural network-based front-end to detect new objects and track 2D bounding boxes between frames. The class label and bounding box is passed to the back-end that updates the object's pose using an unscented Kalman filter (UKF). The measurement posterior is fed back to the 2D tracker to improve robustness. The object's class is identified so a class-specific UKF can be used if custom dynamics and constraints are known. Adapting to track the pose of new classes only requires providing a trained 2D object detector or labeled 2D bounding box data, as well as the approximate size of the objects. The performance of MSL-RAPTOR is first verified on the NOCS-REAL275 dataset, achieving results comparable to RGB-D approaches despite not using depth measurements. When tracking a flying drone from onboard another drone, it outperforms the fastest comparable method in speed by a factor of 3, while giving lower translation and rotation median errors by 66% and 23% respectively.

Haruki Nishimura, Negar Mehr, Adrien Gaidon et al.

Successful robotic operation in stochastic environments relies on accurate characterization of the underlying probability distributions, yet this is often imperfect due to limited knowledge. This work presents a control algorithm that is capable of handling such distributional mismatches. Specifically, we propose a novel nonlinear MPC for distributionally robust control, which plans locally optimal feedback policies against a worst-case distribution within a given KL divergence bound from a Gaussian distribution. Leveraging mathematical equivalence between distributionally robust control and risk-sensitive optimal control, our framework also provides an algorithm to dynamically adjust the risk-sensitivity level online for risk-sensitive control. The benefits of the distributional robustness as well as the automatic risk-sensitivity adjustment are demonstrated in a dynamic collision avoidance scenario where the predictive distribution of human motion is erroneous.

Eric Cristofalo, Eduardo Montijano, Mac Schwager

We present a consensus-based distributed pose graph optimization algorithm for obtaining an estimate of the 3D translation and rotation of each pose in a pose graph, given noisy relative measurements between poses. The algorithm, called GeoD, implements a continuous time distributed consensus protocol to minimize the geodesic pose graph error. GeoD is distributed over the pose graph itself, with a separate computation thread for each node in the graph, and messages are passed only between neighboring nodes in the graph. We leverage tools from Lyapunov theory and multi-agent consensus to prove the convergence of the algorithm. We identify two new consistency conditions sufficient for convergence: pairwise consistency of relative rotation measurements, and minimal consistency of relative translation measurements. GeoD incorporates a simple one step distributed initialization to satisfy both conditions. We demonstrate GeoD on simulated and real world SLAM datasets. We compare to a centralized pose graph optimizer with an optimality certificate (SE-Sync) and a Distributed Gauss-Seidel (DGS) method. On average, GeoD converges 20 times more quickly than DGS to a value with 3.4 times less error when compared to the global minimum provided by SE-Sync. GeoD scales more favorably with graph size than DGS, converging over 100 times faster on graphs larger than 1000 poses. Lastly, we test GeoD on a multi-UAV vision-based SLAM scenario, where the UAVs estimate their pose trajectories in a distributed manner using the relative poses extracted from their on board camera images. We show qualitative performance that is better than either the centralized SE-Sync or the distributed DGS methods.