Daniel Szafir

Maitrey Gramopadhye, Daniel Szafir

Large Language Models (LLMs) trained using massive text datasets have recently shown promise in generating action plans for robotic agents from high level text queries. However, these models typically do not consider the robot's environment, resulting in generated plans that may not actually be executable, due to ambiguities in the planned actions or environmental constraints. In this paper, we propose an approach to generate environmentally-aware action plans that agents are better able to execute. Our approach involves integrating environmental objects and object relations as additional inputs into LLM action plan generation to provide the system with an awareness of its surroundings, resulting in plans where each generated action is mapped to objects present in the scene. We also design a novel scoring function that, along with generating the action steps and associating them with objects, helps the system disambiguate among object instances and take into account their states. We evaluated our approach using the VirtualHome simulator and the ActivityPrograms knowledge base and found that action plans generated from our system had a 310% improvement in executability and a 147% improvement in correctness over prior work. The complete code and a demo of our method is publicly available at https://github.com/hri-ironlab/scene_aware_language_planner.

Yu Fang, Kanchana Ranasinghe, Le Xue et al. · salesforce, stanford

Vision-Language-Action (VLA) models have achieved remarkable progress in robotic manipulation by mapping multimodal observations and instructions directly to actions. However, they typically mimic expert trajectories without predictive motion reasoning, which limits their ability to reason about what actions to take. To address this limitation, we propose joint learning with motion image diffusion, a novel strategy that enhances VLA models with motion reasoning capabilities. Our method extends the VLA architecture with a dual-head design: while the action head predicts action chunks as in vanilla VLAs, an additional motion head, implemented as a Diffusion Transformer (DiT), predicts optical-flow-based motion images that capture future dynamics. The two heads are trained jointly, enabling the shared VLM backbone to learn representations that couple robot control with motion knowledge. This joint learning builds temporally coherent and physically grounded representations without modifying the inference pathway of standard VLAs, thereby maintaining test-time latency. Experiments in both simulation and real-world environments demonstrate that joint learning with motion image diffusion improves the success rate of pi-series VLAs to 97.5% on the LIBERO benchmark and 58.0% on the RoboTwin benchmark, yielding a 23% improvement in real-world performance and validating its effectiveness in enhancing the motion reasoning capability of large-scale VLAs.

Yue Yang, Shuo Cheng, Yu Fang et al.

General-purpose robots must master long-horizon manipulation, defined as tasks involving multiple kinematic structure changes (e.g., attaching or detaching objects) in unstructured environments. While Vision-Language-Action (VLA) models offer the potential to master diverse atomic skills, they struggle with the combinatorial complexity of sequencing them and are prone to cascading failures due to environmental sensitivity. To address these challenges, we propose LiLo-VLA (Linked Local VLA), a modular framework capable of zero-shot generalization to novel long-horizon tasks without ever being trained on them. Our approach decouples transport from interaction: a Reaching Module handles global motion, while an Interaction Module employs an object-centric VLA to process isolated objects of interest, ensuring robustness against irrelevant visual features and invariance to spatial configurations. Crucially, this modularity facilitates robust failure recovery through dynamic replanning and skill reuse, effectively mitigating the cascading errors common in end-to-end approaches. We introduce a 21-task simulation benchmark consisting of two challenging suites: LIBERO-Long++ and Ultra-Long. In these simulations, LiLo-VLA achieves a 69% average success rate, outperforming Pi0.5 by 41% and OpenVLA-OFT by 67%. Furthermore, real-world evaluations across 8 long-horizon tasks demonstrate an average success rate of 85%. Project page: https://yy-gx.github.io/LiLo-VLA/.

Diane Jung, Caleb Escobedo, Noah Liska et al.

Scientists perform diverse manual procedures that are tedious and laborious. Such procedures are considered a bottleneck for modern experimental science, as they consume time and increase burdens in fields including material science and medicine. We employ a user-centered approach to designing a robot-assisted system for dialysis, a common multi-day purification method used in polymer and protein synthesis. Through two usability studies, we obtain participant feedback and revise design requirements to develop the final system that satisfies scientists' needs and has the potential for applications in other experimental workflows. We anticipate that integration of this system into real synthesis procedures in a chemical wet lab will decrease workload on scientists during long experimental procedures and provide an effective approach to designing more systems that have the potential to accelerate scientific discovery and liberate scientists from tedious labor.

Yu Fang, Yuchun Feng, Dong Jing et al.

Vision-Language-Action models (VLAs) promise to ground language instructions in robot control, yet in practice often fail to faithfully follow language. When presented with instructions that lack strong scene-specific supervision, VLAs suffer from counterfactual failures: they act based on vision shortcuts induced by dataset biases, repeatedly executing well-learned behaviors and selecting objects frequently seen during training regardless of language intent. To systematically study it, we introduce LIBERO-CF, the first counterfactual benchmark for VLAs that evaluates language following capability by assigning alternative instructions under visually plausible LIBERO layouts. Our evaluation reveals that counterfactual failures are prevalent yet underexplored across state-of-the-art VLAs. We propose Counterfactual Action Guidance (CAG), a simple yet effective dual-branch inference scheme that explicitly regularizes language conditioning in VLAs. CAG combines a standard VLA policy with a language-unconditioned Vision-Action (VA) module, enabling counterfactual comparison during action selection. This design reduces reliance on visual shortcuts, improves robustness on under-observed tasks, and requires neither additional demonstrations nor modifications to existing architectures or pretrained models. Extensive experiments demonstrate its plug-and-play integration across diverse VLAs and consistent improvements. For example, on LIBERO-CF, CAG improves $π_{0.5}$ by 9.7% in language following accuracy and 3.6% in task success on under-observed tasks using a training-free strategy, with further gains of 15.5% and 8.5%, respectively, when paired with a VA model. In real-world evaluations, CAG reduces counterfactual failures of 9.4% and improves task success by 17.2% on average.

Yu Fang, Yue Yang, Xinghao Zhu et al.

Vision-language-action (VLA) models present a promising paradigm by training policies directly on real robot datasets like Open X-Embodiment. However, the high cost of real-world data collection hinders further data scaling, thereby restricting the generalizability of VLAs. In this paper, we introduce ReBot, a novel real-to-sim-to-real approach for scaling real robot datasets and adapting VLA models to target domains, which is the last-mile deployment challenge in robot manipulation. Specifically, ReBot replays real-world robot trajectories in simulation to diversify manipulated objects (real-to-sim), and integrates the simulated movements with inpainted real-world background to synthesize physically realistic and temporally consistent robot videos (sim-to-real). Our approach has several advantages: 1) it enjoys the benefit of real data to minimize the sim-to-real gap; 2) it leverages the scalability of simulation; and 3) it can generalize a pretrained VLA to a target domain with fully automated data pipelines. Extensive experiments in both simulation and real-world environments show that ReBot significantly enhances the performance and robustness of VLAs. For example, in SimplerEnv with the WidowX robot, ReBot improved the in-domain performance of Octo by 7.2% and OpenVLA by 21.8%, and out-of-domain generalization by 19.9% and 9.4%, respectively. For real-world evaluation with a Franka robot, ReBot increased the success rates of Octo by 17% and OpenVLA by 20%. More information can be found at: https://yuffish.github.io/rebot/

Yue Yang, Linfeng Zhao, Mingyu Ding et al.

Robotics has long sought to develop visual-servoing robots capable of completing previously unseen long-horizon tasks. Hierarchical approaches offer a pathway for achieving this goal by executing skill combinations arranged by a task planner, with each visuomotor skill pre-trained using a specific imitation learning (IL) algorithm. However, even in simple long-horizon tasks like skill chaining, hierarchical approaches often struggle due to a problem we identify as Observation Space Shift (OSS), where the sequential execution of preceding skills causes shifts in the observation space, disrupting the performance of subsequent individually trained skill policies. To validate OSS and evaluate its impact on long-horizon tasks, we introduce BOSS (a Benchmark for Observation Space Shift). BOSS comprises three distinct challenges: "Single Predicate Shift", "Accumulated Predicate Shift", and "Skill Chaining", each designed to assess a different aspect of OSS's negative effect. We evaluated several recent popular IL algorithms on BOSS, including three Behavioral Cloning methods and the Visual Language Action model OpenVLA. Even on the simplest challenge, we observed average performance drops of 67%, 35%, 34%, and 54%, respectively, when comparing skill performance with and without OSS. Additionally, we investigate a potential solution to OSS that scales up the training data for each skill with a larger and more visually diverse set of demonstrations, with our results showing it is not sufficient to resolve OSS. The project page is: https://boss-benchmark.github.io/

Yue Yang, Bryce Ikeda, Gedas Bertasius et al.

Robot Imitation Learning (IL) is a widely used method for training robots to perform manipulation tasks that involve mimicking human demonstrations to acquire skills. However, its practicality has been limited due to its requirement that users be trained in operating real robot arms to provide demonstrations. This paper presents an innovative solution: an Augmented Reality (AR)-assisted framework for demonstration collection, empowering non-roboticist users to produce demonstrations for robot IL using devices like the HoloLens 2. Our framework facilitates scalable and diverse demonstration collection for real-world tasks. We validate our approach with experiments on three classical robotics tasks: reach, push, and pick-and-place. The real robot performs each task successfully while replaying demonstrations collected via AR.

Michael Walker, Thao Phung, Tathagata Chakraborti et al.

Virtual, Augmented, and Mixed Reality for Human-Robot Interaction (VAM-HRI) has been gaining considerable attention in research in recent years. However, the HRI community lacks a set of shared terminology and framework for characterizing aspects of mixed reality interfaces, presenting serious problems for future research. Therefore, it is important to have a common set of terms and concepts that can be used to precisely describe and organize the diverse array of work being done within the field. In this paper, we present a novel taxonomic framework for different types of VAM-HRI interfaces, composed of four main categories of virtual design elements (VDEs). We present and justify our taxonomy and explain how its elements have been developed over the last 30 years as well as the current directions VAM-HRI is headed in the coming decade.

Connor Brooks, Daniel Szafir

In shared control, advances in autonomous robotics are applied to help empower a human user in operating a robotic system. While these systems have been shown to improve efficiency and operation success, users are not always accepting of the new control paradigm produced by working with an assistive controller. This mismatch between performance and acceptance can prevent users from taking advantage of the benefits of shared control systems for robotic operation. To address this mismatch, we develop multiple types of visualizations for improving both the legibility and perceived predictability of assistive controllers, then conduct a user study to evaluate the impact that these visualizations have on user acceptance of shared control systems. Our results demonstrate that shared control visualizations must be designed carefully to be effective, with users requiring visualizations that improve both legibility and predictability of the assistive controller in order to voluntarily relinquish control.

Ryo Suzuki, Hooman Hedayati, Clement Zheng et al.

RoomShift is a room-scale dynamic haptic environment for virtual reality, using a small swarm of robots that can move furniture. RoomShift consists of nine shape-changing robots: Roombas with mechanical scissor lifts. These robots drive beneath a piece of furniture to lift, move and place it. By augmenting virtual scenes with physical objects, users can sit on, lean against, place and otherwise interact with furniture with their whole body; just as in the real world. When the virtual scene changes or users navigate within it, the swarm of robots dynamically reconfigures the physical environment to match the virtual content. We describe the hardware and software implementation, applications in virtual tours and architectural design and interaction techniques.

Hooman Hedayati, Ryo Suzuki1, Daniel Leithinger et al.

We present PufferBot, an aerial robot with an expandable structure that may expand to protect a drone's propellers when the robot is close to obstacles or collocated humans. PufferBot is made of a custom 3D-printed expandable scissor structure, which utilizes a one degree of freedom actuator with rack and pinion mechanism. We propose four designs for the expandable structure, each with unique characterizations for different situations. Finally, we present three motivating scenarios in which PufferBot may extend the utility of existing static propeller guard structures. The supplementary video can be found at: https://youtu.be/XtPepCxWcBg

Connor Brooks, Daniel Szafir

The mental models that humans form of other agents---encapsulating human beliefs about agent goals, intentions, capabilities, and more---create an underlying basis for interaction. These mental models have the potential to affect both the human's decision making during the interaction and the human's subjective assessment of the interaction. In this paper, we surveyed existing methods for modeling how humans view robots, then identified a potential method for improving these estimates through inferring a human's model of a robot agent directly from their actions. Then, we conducted an online study to collect data in a grid-world environment involving humans moving an avatar past a virtual agent. Through our analysis, we demonstrated that participants' action choices leaked information about their mental models of a virtual agent. We conclude by discussing the implications of these findings and the potential for such a method to improve human-robot interactions.

Michael L. Iuzzolino, Michael E. Walker, Daniel Szafir

Robots hold promise in many scenarios involving outdoor use, such as search-and-rescue, wildlife management, and collecting data to improve environment, climate, and weather forecasting. However, autonomous navigation of outdoor trails remains a challenging problem. Recent work has sought to address this issue using deep learning. Although this approach has achieved state-of-the-art results, the deep learning paradigm may be limited due to a reliance on large amounts of annotated training data. Collecting and curating training datasets may not be feasible or practical in many situations, especially as trail conditions may change due to seasonal weather variations, storms, and natural erosion. In this paper, we explore an approach to address this issue through virtual-to-real-world transfer learning using a variety of deep learning models trained to classify the direction of a trail in an image. Our approach utilizes synthetic data gathered from virtual environments for model training, bypassing the need to collect a large amount of real images of the outdoors. We validate our approach in three main ways. First, we demonstrate that our models achieve classification accuracies upwards of 95% on our synthetic data set. Next, we utilize our classification models in the control system of a simulated robot to demonstrate feasibility. Finally, we evaluate our models on real-world trail data and demonstrate the potential of virtual-to-real-world transfer learning.

Steve McGuire, Christoffer Heckman, Daniel Szafir et al.

Determining extrinsic calibration parameters is a necessity in any robotic system composed of actuators and cameras. Once a system is outside the lab environment, parameters must be determined without relying on outside artifacts such as calibration targets. We propose a method that relies on structured motion of an observed arm to recover extrinsic calibration parameters. Our method combines known arm kinematics with observations of conics in the image plane to calculate maximum-likelihood estimates for calibration extrinsics. This method is validated in simulation and tested against a real-world model, yielding results consistent with ruler-based estimates. Our method shows promise for estimating the pose of a camera relative to an articulated arm's end effector without requiring tedious measurements or external artifacts. Index Terms: robotics, hand-eye problem, self-calibration, structure from motion