Matthew Johnson‐Roberson

Quanting Xie, So Yeon Min, Pengliang Ji et al. · cmu

There is no limit to how much a robot might explore and learn, but all of that knowledge needs to be searchable and actionable. Within language research, retrieval augmented generation (RAG) has become the workhorse of large-scale non-parametric knowledge; however, existing techniques do not directly transfer to the embodied domain, which is multimodal, where data is highly correlated, and perception requires abstraction. To address these challenges, we introduce Embodied-RAG, a framework that enhances the foundational model of an embodied agent with a non-parametric memory system capable of autonomously constructing hierarchical knowledge for both navigation and language generation. Embodied-RAG handles a full range of spatial and semantic resolutions across diverse environments and query types, whether for a specific object or a holistic description of ambiance. At its core, Embodied-RAG's memory is structured as a semantic forest, storing language descriptions at varying levels of detail. This hierarchical organization allows the system to efficiently generate context-sensitive outputs across different robotic platforms. We demonstrate that Embodied-RAG effectively bridges RAG to the robotics domain, successfully handling over 250 explanation and navigation queries across kilometer-level environments, highlighting its promise as a general-purpose non-parametric system for embodied agents.

Quanting Xie, Tianyi Zhang, Kedi Xu et al. · cmu

Robots should exist anywhere humans do: indoors, outdoors, and even unmapped environments. In contrast, the focus of recent advancements in Object Goal Navigation(OGN) has targeted navigating in indoor environments by leveraging spatial and semantic cues that do not generalize outdoors. While these contributions provide valuable insights into indoor scenarios, the broader spectrum of real-world robotic applications often extends to outdoor settings. As we transition to the vast and complex terrains of outdoor environments, new challenges emerge. Unlike the structured layouts found indoors, outdoor environments lack clear spatial delineations and are riddled with inherent semantic ambiguities. Despite this, humans navigate with ease because we can reason about the unseen. We introduce a new task OUTDOOR, a new mechanism for Large Language Models (LLMs) to accurately hallucinate possible futures, and a new computationally aware success metric for pushing research forward in this more complex domain. Additionally, we show impressive results on both a simulated drone and physical quadruped in outdoor environments. Our agent has no premapping and our formalism outperforms naive LLM-based approaches

Tianyi Zhang, Matthew Johnson-Roberson

Underwater imagery often exhibits distorted coloration as a result of light-water interactions, which complicates the study of benthic environments in marine biology and geography. In this research, we propose an algorithm to restore the true color (albedo) in underwater imagery by jointly learning the effects of the medium and neural scene representations. Our approach models water effects as a combination of light attenuation with distance and backscattered light. The proposed neural scene representation is based on a neural reflectance field model, which learns albedos, normals, and volume densities of the underwater environment. We introduce a logistic regression model to separate water from the scene and apply distinct light physics during training. Our method avoids the need to estimate complex backscatter effects in water by employing several approximations, enhancing sampling efficiency and numerical stability during training. The proposed technique integrates underwater light effects into a volume rendering framework with end-to-end differentiability. Experimental results on both synthetic and real-world data demonstrate that our method effectively restores true color from underwater imagery, outperforming existing approaches in terms of color consistency.

Ming-Yuan Yu, Ram Vasudevan, Matthew Johnson-Roberson

LiDARs have been widely adopted to modern self-driving vehicles, providing 3D information of the scene and surrounding objects. However, adverser weather conditions still pose significant challenges to LiDARs since point clouds captured during snowfall can easily be corrupted. The resulting noisy point clouds degrade downstream tasks such as mapping. Existing works in de-noising point clouds corrupted by snow are based on nearest-neighbor search, and thus do not scale well with modern LiDARs which usually capture $100k$ or more points at 10Hz. In this paper, we introduce an unsupervised de-noising algorithm, LiSnowNet, running 52$\times$ faster than the state-of-the-art methods while achieving superior performance in de-noising. Unlike previous methods, the proposed algorithm is based on a deep convolutional neural network and can be easily deployed to hardware accelerators such as GPUs. In addition, we demonstrate how to use the proposed method for mapping even with corrupted point clouds.

Alexandra Carlson, Manikandasriram Srinivasan Ramanagopal, Nathan Tseng et al.

Recent advances in neural radiance fields (NeRFs) achieve state-of-the-art novel view synthesis and facilitate dense estimation of scene properties. However, NeRFs often fail for large, unbounded scenes that are captured under very sparse views with the scene content concentrated far away from the camera, as is typical for field robotics applications. In particular, NeRF-style algorithms perform poorly: (1) when there are insufficient views with little pose diversity, (2) when scenes contain saturation and shadows, and (3) when finely sampling large unbounded scenes with fine structures becomes computationally intensive. This paper proposes CLONeR, which significantly improves upon NeRF by allowing it to model large outdoor driving scenes that are observed from sparse input sensor views. This is achieved by decoupling occupancy and color learning within the NeRF framework into separate Multi-Layer Perceptrons (MLPs) trained using LiDAR and camera data, respectively. In addition, this paper proposes a novel method to build differentiable 3D Occupancy Grid Maps (OGM) alongside the NeRF model, and leverage this occupancy grid for improved sampling of points along a ray for volumetric rendering in metric space. Through extensive quantitative and qualitative experiments on scenes from the KITTI dataset, this paper demonstrates that the proposed method outperforms state-of-the-art NeRF models on both novel view synthesis and dense depth prediction tasks when trained on sparse input data.

Weiming Zhi, Tianyi Zhang, Matthew Johnson-Roberson

Learning for Demonstration (LfD) enables robots to acquire new skills by imitating expert demonstrations, allowing users to communicate their instructions in an intuitive manner. Recent progress in LfD often relies on kinesthetic teaching or teleoperation as the medium for users to specify the demonstrations. Kinesthetic teaching requires physical handling of the robot, while teleoperation demands proficiency with additional hardware. This paper introduces an alternative paradigm for LfD called Diagrammatic Teaching. Diagrammatic Teaching aims to teach robots novel skills by prompting the user to sketch out demonstration trajectories on 2D images of the scene, these are then synthesised as a generative model of motion trajectories in 3D task space. Additionally, we present the Ray-tracing Probabilistic Trajectory Learning (RPTL) framework for Diagrammatic Teaching. RPTL extracts time-varying probability densities from the 2D sketches, applies ray-tracing to find corresponding regions in 3D Cartesian space, and fits a probabilistic model of motion trajectories to these regions. New motion trajectories, which mimic those sketched by the user, can then be generated from the probabilistic model. We empirically validate our framework both in simulation and on real robots, which include a fixed-base manipulator and a quadruped-mounted manipulator.

Junming Zhang, Haomeng Zhang, Ram Vasudevan et al.

Most real-world 3D measurements from depth sensors are incomplete, and to address this issue the point cloud completion task aims to predict the complete shapes of objects from partial observations. Previous works often adapt an encoder-decoder architecture, where the encoder is trained to extract embeddings that are used as inputs to generate predictions from the decoder. However, the learned embeddings have sparse distribution in the feature space, which leads to worse generalization results during testing. To address these problems, this paper proposes a hyperspherical module, which transforms and normalizes embeddings from the encoder to be on a unit hypersphere. With the proposed module, the magnitude and direction of the output hyperspherical embedding are decoupled and only the directional information is optimized. We theoretically analyze the hyperspherical embedding and show that it enables more stable training with a wider range of learning rates and more compact embedding distributions. Experiment results show consistent improvement of point cloud completion in both single-task and multi-task learning, which demonstrates the effectiveness of the proposed method.

Weiming Zhi, Tianyi Zhang, Matthew Johnson-Roberson

Diagrammatic Teaching is a paradigm for robots to acquire novel skills, whereby the user provides 2D sketches over images of the scene to shape the robot's motion. In this work, we tackle the problem of teaching a robot to approach a surface and then follow cyclic motion on it, where the cycle of the motion can be arbitrarily specified by a single user-provided sketch over an image from the robot's camera. Accordingly, we contribute the Stable Diffeomorphic Diagrammatic Teaching (SDDT) framework. SDDT models the robot's motion as an Orbitally Asymptotically Stable (O.A.S.) dynamical system that learns to stablize based on a single diagrammatic sketch provided by the user. This is achieved by applying a \emph{diffeomorphism}, i.e. a differentiable and invertible function, to morph a known O.A.S. system. The parameterised diffeomorphism is then optimised with respect to the Hausdorff distance between the limit cycle of our modelled system and the sketch, to produce the desired robot motion. We provide novel theoretical insight into the behaviour of the optimised system and also empirically evaluate SDDT, both in simulation and on a quadruped with a mounted 6-DOF manipulator. Results show that we can diagrammatically teach complex cyclic motion patterns with a high degree of accuracy.

Tianyi Zhang, Weiming Zhi, Kaining Huang et al.

Water caustics are commonly observed in seafloor imaging data from shallow-water areas. Traditional methods that remove caustic patterns from images often rely on 2D filtering or pre-training on an annotated dataset, hindering the performance when generalizing to real-world seafloor data with 3D structures. In this paper, we present a novel method Recurrent Gaussian Splatting (RecGS), which takes advantage of today's photorealistic 3D reconstruction technology, 3DGS, to separate caustics from seafloor imagery. With a sequence of images taken by an underwater robot, we build 3DGS recurrently and decompose the caustic with low-pass filtering in each iteration. In the experiments, we analyze and compare with different methods, including joint optimization, 2D filtering, and deep learning approaches. The results show that our method can effectively separate the caustic from the seafloor, improving the visual appearance, and can be potentially applied on more problems with inconsistent illumination.

Xiaohao Xu, Tianyi Zhang, Sibo Wang et al.

Robustness is a crucial factor for the successful deployment of robots in unstructured environments, particularly in the domain of Simultaneous Localization and Mapping (SLAM). Simulation-based benchmarks have emerged as a highly scalable approach for robustness evaluation compared to real-world data collection. However, crafting a challenging and controllable noisy world with diverse perturbations remains relatively under-explored. To this end, we propose a novel, customizable pipeline for noisy data synthesis, aimed at assessing the resilience of multi-modal SLAM models against various perturbations. This pipeline incorporates customizable hardware setups, software components, and perturbed environments. In particular, we introduce comprehensive perturbation taxonomy along with a perturbation composition toolbox, allowing the transformation of clean simulations into challenging noisy environments. Utilizing the pipeline, we instantiate the Robust-SLAM benchmark, which includes diverse perturbation types, to evaluate the risk tolerance of existing advanced multi-modal SLAM models. Our extensive analysis uncovers the susceptibilities of existing SLAM models to real-world disturbance, despite their demonstrated accuracy in standard benchmarks. Our perturbation synthesis toolbox, SLAM robustness evaluation pipeline, and Robust-SLAM benchmark will be made publicly available at https://github.com/Xiaohao-Xu/SLAM-under-Perturbation/.

Weiming Zhi, Haozhan Tang, Tianyi Zhang et al.

Humans have the remarkable ability to use held objects as tools to interact with their environment. For this to occur, humans internally estimate how hand movements affect the object's movement. We wish to endow robots with this capability. We contribute methodology to jointly estimate the geometry and pose of objects grasped by a robot, from RGB images captured by an external camera. Notably, our method transforms the estimated geometry into the robot's coordinate frame, while not requiring the extrinsic parameters of the external camera to be calibrated. Our approach leverages 3D foundation models, large models pre-trained on huge datasets for 3D vision tasks, to produce initial estimates of the in-hand object. These initial estimations do not have physically correct scales and are in the camera's frame. Then, we formulate, and efficiently solve, a coordinate-alignment problem to recover accurate scales, along with a transformation of the objects to the coordinate frame of the robot. Forward kinematics mappings can subsequently be defined from the manipulator's joint angles to specified points on the object. These mappings enable the estimation of points on the held object at arbitrary configurations, enabling robot motion to be designed with respect to coordinates on the grasped objects. We empirically evaluate our approach on a robot manipulator holding a diverse set of real-world objects.

Haozhan Tang, Tianyi Zhang, Oliver Kroemer et al.

Robots operating in unstructured environments often require accurate and consistent object-level representations. This typically requires segmenting individual objects from the robot's surroundings. While recent large models such as Segment Anything (SAM) offer strong performance in 2D image segmentation. These advances do not translate directly to performance in the physical 3D world, where they often over-segment objects and fail to produce consistent mask correspondences across views. In this paper, we present GraphSeg, a framework for generating consistent 3D object segmentations from a sparse set of 2D images of the environment without any depth information. GraphSeg adds edges to graphs and constructs dual correspondence graphs: one from 2D pixel-level similarities and one from inferred 3D structure. We formulate segmentation as a problem of edge addition, then subsequent graph contraction, which merges multiple 2D masks into unified object-level segmentations. We can then leverage \emph{3D foundation models} to produce segmented 3D representations. GraphSeg achieves robust segmentation with significantly fewer images and greater accuracy than prior methods. We demonstrate state-of-the-art performance on tabletop scenes and show that GraphSeg enables improved performance on downstream robotic manipulation tasks. Code available at https://github.com/tomtang502/graphseg.git.

Yafei Hu, Quanting Xie, Vidhi Jain et al. · cmu

Building general-purpose robots that operate seamlessly in any environment, with any object, and utilizing various skills to complete diverse tasks has been a long-standing goal in Artificial Intelligence. However, as a community, we have been constraining most robotic systems by designing them for specific tasks, training them on specific datasets, and deploying them within specific environments. These systems require extensively-labeled data and task-specific models. When deployed in real-world scenarios, such systems face several generalization issues and struggle to remain robust to distribution shifts. Motivated by the impressive open-set performance and content generation capabilities of web-scale, large-capacity pre-trained models (i.e., foundation models) in research fields such as Natural Language Processing (NLP) and Computer Vision (CV), we devote this survey to exploring (i) how these existing foundation models from NLP and CV can be applied to the field of general-purpose robotics, and also exploring (ii) what a robotics-specific foundation model would look like. We begin by providing a generalized formulation of how foundation models are used in robotics, and the fundamental barriers to making generalist robots universally applicable. Next, we establish a taxonomy to discuss current work exploring ways to leverage existing foundation models for robotics and develop ones catered to robotics. Finally, we discuss key challenges and promising future directions in using foundation models for enabling general-purpose robotic systems. We encourage readers to view our living GitHub repository 2 of resources, including papers reviewed in this survey, as well as related projects and repositories for developing foundation models for robotics.

Xiaohao Xu, Tianyi Zhang, Sibo Wang et al.

Embodied agents require robust navigation systems to operate in unstructured environments, making the robustness of Simultaneous Localization and Mapping (SLAM) models critical to embodied agent autonomy. While real-world datasets are invaluable, simulation-based benchmarks offer a scalable approach for robustness evaluations. However, the creation of a challenging and controllable noisy world with diverse perturbations remains under-explored. To this end, we propose a novel, customizable pipeline for noisy data synthesis, aimed at assessing the resilience of multi-modal SLAM models against various perturbations. The pipeline comprises a comprehensive taxonomy of sensor and motion perturbations for embodied multi-modal (specifically RGB-D) sensing, categorized by their sources and propagation order, allowing for procedural composition. We also provide a toolbox for synthesizing these perturbations, enabling the transformation of clean environments into challenging noisy simulations. Utilizing the pipeline, we instantiate the large-scale Noisy-Replica benchmark, which includes diverse perturbation types, to evaluate the risk tolerance of existing advanced RGB-D SLAM models. Our extensive analysis uncovers the susceptibilities of both neural (NeRF and Gaussian Splatting -based) and non-neural SLAM models to disturbances, despite their demonstrated accuracy in standard benchmarks. Our code is publicly available at https://github.com/Xiaohao-Xu/SLAM-under-Perturbation.

Gideon Billings, Eduardo Iscar, Matthew Johnson-Roberson

The design of optical systems for underwater vehicles is a complex process where the selection of cameras, lenses, housings, and operational parameters greatly influence the performance of the complete system. Determining the correct combination of components and parameters for a given set of operational requirements is currently a process based on trial and error as well as the specialized knowledge and experience of the designer. In this paper, we introduce an open-source tool for the parametric exploration of the design space of underwater optical systems and review the most significant underwater light effects with the corresponding models to estimate the response and performance of the complete imaging system.

Tianyi Zhang, Kaining Huang, Weiming Zhi et al.

Humans have the remarkable ability to construct consistent mental models of an environment, even under limited or varying levels of illumination. We wish to endow robots with this same capability. In this paper, we tackle the challenge of constructing a photorealistic scene representation under poorly illuminated conditions and with a moving light source. We approach the task of modeling illumination as a learning problem, and utilize the developed illumination model to aid in scene reconstruction. We introduce an innovative framework that uses a data-driven approach, Neural Light Simulators (NeLiS), to model and calibrate the camera-light system. Furthermore, we present DarkGS, a method that applies NeLiS to create a relightable 3D Gaussian scene model capable of real-time, photorealistic rendering from novel viewpoints. We show the applicability and robustness of our proposed simulator and system in a variety of real-world environments.

Xiaohao Xu, Tianyi Zhang, Shibo Zhao et al.

We aim to redefine robust ego-motion estimation and photorealistic 3D reconstruction by addressing a critical limitation: the reliance on noise-free data in existing models. While such sanitized conditions simplify evaluation, they fail to capture the unpredictable, noisy complexities of real-world environments. Dynamic motion, sensor imperfections, and synchronization perturbations lead to sharp performance declines when these models are deployed in practice, revealing an urgent need for frameworks that embrace and excel under real-world noise. To bridge this gap, we tackle three core challenges: scalable data generation, comprehensive benchmarking, and model robustness enhancement. First, we introduce a scalable noisy data synthesis pipeline that generates diverse datasets simulating complex motion, sensor imperfections, and synchronization errors. Second, we leverage this pipeline to create Robust-Ego3D, a benchmark rigorously designed to expose noise-induced performance degradation, highlighting the limitations of current learning-based methods in ego-motion accuracy and 3D reconstruction quality. Third, we propose Correspondence-guided Gaussian Splatting (CorrGS), a novel test-time adaptation method that progressively refines an internal clean 3D representation by aligning noisy observations with rendered RGB-D frames from clean 3D map, enhancing geometric alignment and appearance restoration through visual correspondence. Extensive experiments on synthetic and real-world data demonstrate that CorrGS consistently outperforms prior state-of-the-art methods, particularly in scenarios involving rapid motion and dynamic illumination.

Weiming Zhi, Haozhan Tang, Tianyi Zhang et al.

Representing the environment is a central challenge in robotics, and is essential for effective decision-making. Traditionally, before capturing images with a manipulator-mounted camera, users need to calibrate the camera using a specific external marker, such as a checkerboard or AprilTag. However, recent advances in computer vision have led to the development of \emph{3D foundation models}. These are large, pre-trained neural networks that can establish fast and accurate multi-view correspondences with very few images, even in the absence of rich visual features. This paper advocates for the integration of 3D foundation models into scene representation approaches for robotic systems equipped with manipulator-mounted RGB cameras. Specifically, we propose the Joint Calibration and Representation (JCR) method. JCR uses RGB images, captured by a manipulator-mounted camera, to simultaneously construct an environmental representation and calibrate the camera relative to the robot's end-effector, in the absence of specific calibration markers. The resulting 3D environment representation is aligned with the robot's coordinate frame and maintains physically accurate scales. We demonstrate that JCR can build effective scene representations using a low-cost RGB camera attached to a manipulator, without prior calibration.

Weiming Zhi, Ziyong Ma, Tianyi Zhang et al.

Autonomous robots typically need to construct representations of their surroundings and adapt their motions to the geometry of their environment. Here, we tackle the problem of constructing a policy model for collision-free motion generation, consistent with the environment, from a single input RGB image. Extracting 3D structures from a single image often involves monocular depth estimation. Developments in depth estimation have given rise to large pre-trained models such as DepthAnything. However, using outputs of these models for downstream motion generation is challenging due to frustum-shaped errors that arise. Instead, we propose a framework known as Video-Generation Environment Representation (VGER), which leverages the advances of large-scale video generation models to generate a moving camera video conditioned on the input image. Frames of this video, which form a multiview dataset, are then input into a pre-trained 3D foundation model to produce a dense point cloud. We then introduce a multi-scale noise approach to train an implicit representation of the environment structure and build a motion generation model that complies with the geometry of the representation. We extensively evaluate VGER over a diverse set of indoor and outdoor environments. We demonstrate its ability to produce smooth motions that account for the captured geometry of a scene, all from a single RGB input image.

Herbert Wright, Weiming Zhi, Martin Matak et al. · nvidia

Constructing 3D representations of object geometry is critical for many robotics tasks, particularly manipulation problems. These representations must be built from potentially noisy partial observations. In this work, we focus on the problem of reconstructing a multi-object scene from a single RGBD image using a fixed camera. Traditional scene representation methods generally cannot infer the geometry of unobserved regions of the objects in the image. Attempts have been made to leverage deep learning to train on a dataset of known objects and representations, and then generalize to new observations. However, this can be brittle to noisy real-world observations and objects not contained in the dataset, and do not provide well-calibrated reconstruction confidences. We propose BRRP, a reconstruction method that leverages preexisting mesh datasets to build an informative prior during robust probabilistic reconstruction. We introduce the concept of a retrieval-augmented prior, where we retrieve relevant components of our prior distribution from a database of objects during inference. The resulting prior enables estimation of the geometry of occluded portions of the in-scene objects. Our method produces a distribution over object shape that can be used for reconstruction and measuring uncertainty. We evaluate our method in both simulated scenes and in the real world. We demonstrate the robustness of our method against deep learning-only approaches while being more accurate than a method without an informative prior. Through real-world experiments, we particularly highlight the capability of BRRP to enable successful dexterous manipulation in clutter.

William Barron, Xiaoxiang Dong, Matthew Johnson-Roberson et al.

Teaching robots novel behaviors typically requires motion demonstrations via teleoperation or kinaesthetic teaching, that is, physically guiding the robot. While recent work has explored using human sketches to specify desired behaviors, data collection remains cumbersome, and demonstration datasets are difficult to scale. In this paper, we introduce an alternative paradigm, Learning from Cross-Modal Instructions, where robots are shaped by demonstrations in the form of rough annotations, which can contain free-form text labels, and are used in lieu of physical motion. We introduce the CrossInstruct framework, which integrates cross-modal instructions as examples into the context input to a foundational vision-language model (VLM). The VLM then iteratively queries a smaller, fine-tuned model, and synthesizes the desired motion over multiple 2D views. These are then subsequently fused into a coherent distribution over 3D motion trajectories in the robot's workspace. By incorporating the reasoning of the large VLM with a fine-grained pointing model, CrossInstruct produces executable robot behaviors that generalize beyond the environment of in the limited set of instruction examples. We then introduce a downstream reinforcement learning pipeline that leverages CrossInstruct outputs to efficiently learn policies to complete fine-grained tasks. We rigorously evaluate CrossInstruct on benchmark simulation tasks and real hardware, demonstrating effectiveness without additional fine-tuning and providing a strong initialization for policies subsequently refined via reinforcement learning.

Xiaoxiang Dong, Matthew Johnson-Roberson, Weiming Zhi

Learning from human video demonstrations offers a scalable alternative to teleoperation or kinesthetic teaching, but poses challenges for robot manipulators due to embodiment differences and joint feasibility constraints. We address this problem by proposing the Joint Flow Trajectory Optimization (JFTO) framework for grasp pose generation and object trajectory imitation under the video-based Learning-from-Demonstration (LfD) paradigm. Rather than directly imitating human hand motions, our method treats demonstrations as object-centric guides, balancing three objectives: (i) selecting a feasible grasp pose, (ii) generating object trajectories consistent with demonstrated motions, and (iii) ensuring collision-free execution within robot kinematics. To capture the multimodal nature of demonstrations, we extend flow matching to $\SE(3)$ for probabilistic modeling of object trajectories, enabling density-aware imitation that avoids mode collapse. The resulting optimization integrates grasp similarity, trajectory likelihood, and collision penalties into a unified differentiable objective. We validate our approach in both simulation and real-world experiments across diverse real-world manipulation tasks.

Wei-Teng Chu, Tianyi Zhang, Matthew Johnson-Roberson et al.

Implicit representations have been widely applied in robotics for obstacle avoidance and path planning. In this paper, we explore the problem of constructing an implicit distance representation from a single image. Past methods for implicit surface reconstruction, such as \emph{NeuS} and its variants generally require a large set of multi-view images as input, and require long training times. In this work, we propose Fast Image-to-Neural Surface (FINS), a lightweight framework that can reconstruct high-fidelity surfaces and SDF fields based on a single or a small set of images. FINS integrates a multi-resolution hash grid encoder with lightweight geometry and color heads, making the training via an approximate second-order optimizer highly efficient and capable of converging within a few seconds. Additionally, we achieve the construction of a neural surface requiring only a single RGB image, by leveraging pre-trained foundation models to estimate the geometry inherent in the image. Our experiments demonstrate that under the same conditions, our method outperforms state-of-the-art baselines in both convergence speed and accuracy on surface reconstruction and SDF field estimation. Moreover, we demonstrate the applicability of FINS for robot surface following tasks and show its scalability to a variety of benchmark datasets.

Haozhan Tang, Tianyi Zhang, Matthew Johnson-Roberson et al.

Robot manipulation, especially bimanual manipulation, often requires setting up multiple cameras on multiple robot manipulators. Before robot manipulators can generate motion or even build representations of their environments, the cameras rigidly mounted to the robot need to be calibrated. Camera calibration is a cumbersome process involving collecting a set of images, with each capturing a pre-determined marker. In this work, we introduce the Bi-Manual Joint Calibration and Representation Framework (Bi-JCR). Bi-JCR enables multiple robot manipulators, each with cameras mounted, to circumvent taking images of calibration markers. By leveraging 3D foundation models for dense, marker-free multi-view correspondence, Bi-JCR jointly estimates: (i) the extrinsic transformation from each camera to its end-effector, (ii) the inter-arm relative poses between manipulators, and (iii) a unified, scale-consistent 3D representation of the shared workspace, all from the same captured RGB image sets. The representation, jointly constructed from images captured by cameras on both manipulators, lives in a common coordinate frame and supports collision checking and semantic segmentation to facilitate downstream bimanual coordination tasks. We empirically evaluate the robustness of Bi-JCR on a variety of tabletop environments, and demonstrate its applicability on a variety of downstream tasks.

Gideon Billings, Matthew Walter, Oscar Pizarro et al.

In this report, we present the system design, operational strategy, and results of coordinated multi-vehicle field demonstrations of autonomous marine robotic technologies in search-for-life missions within the Pacific shelf margin of Costa Rica and the Santorini-Kolumbo caldera complex, which serve as analogs to environments that may exist in oceans beyond Earth. This report focuses on the automation of ROV manipulator operations for targeted biological sample-collection-and-return from the seafloor. In the context of future extraterrestrial exploration missions to ocean worlds, an ROV is an analog to a planetary lander, which must be capable of high-level autonomy. Our field trials involve two underwater vehicles, the SuBastian ROV and the Nereid Under Ice (NUI) hybrid ROV for mixed initiative (i.e., teleoperated or autonomous) missions, both equipped 7-DoF hydraulic manipulators. We describe an adaptable, hardware-independent computer vision architecture that enables high-level automated manipulation. The vision system provides a 3D understanding of the workspace to inform manipulator motion planning in complex unstructured environments. We demonstrate the effectiveness of the vision system and control framework through field trials in increasingly challenging environments, including the automated collection and return of biological samples from within the active undersea volcano, Kolumbo. Based on our experiences in the field, we discuss the performance of our system and identify promising directions for future research.

Gideon Billings, Richard Camilli, Matthew Johnson-Roberson

This paper presents a novel visual feature based scene mapping method for underwater vehicle manipulator systems (UVMSs), with specific emphasis on robust mapping in natural seafloor environments. Our method uses GPU accelerated SIFT features in a graph optimization framework to build a feature map. The map scale is constrained by features from a vehicle mounted stereo camera, and we exploit the dynamic positioning capability of the manipulator system by fusing features from a wrist mounted fisheye camera into the map to extend it beyond the limited viewpoint of the vehicle mounted cameras. Our hybrid SLAM method is evaluated on challenging image sequences collected with a UVMS in natural deep seafloor environments of the Costa Rican continental shelf margin, and we also evaluate the stereo only mode on a shallow reef survey dataset. Results on these datasets demonstrate the high accuracy of our system and suitability for operating in diverse and natural seafloor environments. We also contribute these datasets for public use.

Tianyi Zhang, Matthew Johnson-Roberson

Robot localization remains a challenging task in GPS denied environments. State estimation approaches based on local sensors, e.g. cameras or IMUs, are drifting-prone for long-range missions as error accumulates. In this study, we aim to address this problem by localizing image observations in a 2D multi-modal geospatial map. We introduce the cross-scale dataset and a methodology to produce additional data from cross-modality sources. We propose a framework that learns cross-scale visual representations without supervision. Experiments are conducted on data from two different domains, underwater and aerial. In contrast to existing studies in cross-view image geo-localization, our approach a) performs better on smaller-scale multi-modal maps; b) is more computationally efficient for real-time applications; c) can serve directly in concert with state estimation pipelines.

Cyrus Anderson, Ram Vasudevan, Matthew Johnson-Roberson

Highway driving invariably combines high speeds with the need to interact closely with other drivers. Prediction methods enable autonomous vehicles (AVs) to anticipate drivers' future trajectories and plan accordingly. Kinematic methods for prediction have traditionally ignored the presence of other drivers, or made predictions only for a limited set of scenarios. Data-driven approaches fill this gap by learning from large datasets to predict trajectories in general scenarios. While they achieve high accuracy, they also lose the interpretability and tools for model validation enjoyed by kinematic methods. This letter proposes a novel kinematic model to describe car-following and lane change behavior, and extends it to predict trajectories in general scenarios. Experiments on highway datasets under varied sensing conditions demonstrate that the proposed method outperforms state-of-the-art methods.

Junming Zhang, Ming-Yuan Yu, Ram Vasudevan et al.

Point cloud analysis is an area of increasing interest due to the development of 3D sensors that are able to rapidly measure the depth of scenes accurately. Unfortunately, applying deep learning techniques to perform point cloud analysis is non-trivial due to the inability of these methods to generalize to unseen rotations. To address this limitation, one usually has to augment the training data, which can lead to extra computation and require larger model complexity. This paper proposes a new neural network called the Aligned Edge Convolutional Neural Network (AECNN) that learns a feature representation of point clouds relative to Local Reference Frames (LRFs) to ensure invariance to rotation. In particular, features are learned locally and aligned with respect to the LRF of an automatically computed reference point. The proposed approach is evaluated on point cloud classification and part segmentation tasks. This paper illustrates that the proposed technique outperforms a variety of state of the art approaches (even those trained on augmented datasets) in terms of robustness to rotation without requiring any additional data augmentation.

Yu Yao, Ella Atkins, Matthew Johnson-Roberson et al.

Pedestrian trajectory prediction is an essential task in robotic applications such as autonomous driving and robot navigation. State-of-the-art trajectory predictors use a conditional variational autoencoder (CVAE) with recurrent neural networks (RNNs) to encode observed trajectories and decode multi-modal future trajectories. This process can suffer from accumulated errors over long prediction horizons (>=2 seconds). This paper presents BiTraP, a goal-conditioned bi-directional multi-modal trajectory prediction method based on the CVAE. BiTraP estimates the goal (end-point) of trajectories and introduces a novel bi-directional decoder to improve longer-term trajectory prediction accuracy. Extensive experiments show that BiTraP generalizes to both first-person view (FPV) and bird's-eye view (BEV) scenarios and outperforms state-of-the-art results by ~10-50%. We also show that different choices of non-parametric versus parametric target models in the CVAE directly influence the predicted multi-modal trajectory distributions. These results provide guidance on trajectory predictor design for robotic applications such as collision avoidance and navigation systems.

Manikandasriram Srinivasan Ramanagopal, Zixu Zhang, Ram Vasudevan et al.

Uncooled microbolometers can enable robots to see in the absence of visible illumination by imaging the "heat" radiated from the scene. Despite this ability to see in the dark, these sensors suffer from significant motion blur. This has limited their application on robotic systems. As described in this paper, this motion blur arises due to the thermal inertia of each pixel. This has meant that traditional motion deblurring techniques, which rely on identifying an appropriate spatial blur kernel to perform spatial deconvolution, are unable to reliably perform motion deblurring on thermal camera images. To address this problem, this paper formulates reversing the effect of thermal inertia at a single pixel as a Least Absolute Shrinkage and Selection Operator (LASSO) problem which we can solve rapidly using a quadratic programming solver. By leveraging sparsity and a high frame rate, this pixel-wise LASSO formulation is able to recover motion deblurred frames of thermal videos without using any spatial information. To compare its quality against state-of-the-art visible camera based deblurring methods, this paper evaluated the performance of a family of pre-trained object detectors on a set of images restored by different deblurring algorithms. All evaluated object detectors performed systematically better on images restored by the proposed algorithm rather than any other tested, state-of-the-art methods.

Cyrus Anderson, Ram Vasudevan, Matthew Johnson-Roberson

Pedestrians and drivers interact closely in a wide range of environments. Autonomous vehicles (AVs) correspondingly face the need to predict pedestrians' future trajectories in these same environments. Traditional model-based prediction methods have been limited to making predictions in highly structured scenes with signalized intersections, marked crosswalks, or curbs. Deep learning methods have instead leveraged datasets to learn predictive features that generalize across scenes, at the cost of model interpretability. This paper aims to achieve both widely applicable and interpretable predictions by proposing a risk-based attention mechanism to learn when pedestrians yield, and a model of vehicle influence to learn how yielding affects motion. A novel probabilistic method, Off the Sidewalk Predictions (OSP), uses these to achieve accurate predictions in both shared spaces and traditional scenes. Experiments on urban datasets demonstrate that the realtime method achieves state-of-the-art performance.

Gideon Billings, Matthew Johnson-Roberson

There has been much recent interest in deep learning methods for monocular image based object pose estimation. While object pose estimation is an important problem for autonomous robot interaction with the physical world, and the application space for monocular-based methods is expansive, there has been little work on applying these methods with fisheye imaging systems. Also, little exists in the way of annotated fisheye image datasets on which these methods can be developed and tested. The research landscape is even more sparse for object detection methods applied in the underwater domain, fisheye image based or otherwise. In this work, we present a novel framework for adapting a ROI-based 6D object pose estimation method to work on full fisheye images. The method incorporates the gnomic projection of regions of interest from an intermediate spherical image representation to correct for the fisheye distortions. Further, we contribute a fisheye image dataset, called UWHandles, collected in natural underwater environments, with 6D object pose and 2D bounding box annotations.

Patrick Holmes, Shreyas Kousik, Bohao Zhang et al.

For robotic arms to operate in arbitrary environments, especially near people, it is critical to certify the safety of their motion planning algorithms. However, there is often a trade-off between safety and real-time performance; one can either carefully design safe plans, or rapidly generate potentially-unsafe plans. This work presents a receding-horizon, real-time trajectory planner with safety guarantees, called ARMTD (Autonomous Reachability-based Manipulator Trajectory Design). The method first computes (offline) a reachable set of parameterized trajectories for each joint of an arm. Each trajectory includes a fail-safe maneuver (braking to a stop). At runtime, in each receding-horizon planning iteration, ARMTD constructs a parameterized reachable set of the full arm in workspace and intersects it with obstacles to generate sub-differentiable, provably-conservative collision-avoidance constraints on the trajectory parameters. ARMTD then performs trajectory optimization over the parameters, subject to these constraints. On a 6 degree-of-freedom arm, ARMTD outperforms CHOMP in simulation, never crashes, and completes a variety of real-time planning tasks on hardware.

Jinsun Liu, Pengcheng Zhao, Zhenyu Gan et al.

Online control design using a high-fidelity, full-order model for a bipedal robot can be challenging due to the size of the state space of the model. A commonly adopted solution to overcome this challenge is to approximate the full-order model (anchor) with a simplified, reduced-order model (template), while performing control synthesis. Unfortunately it is challenging to make formal guarantees about the safety of an anchor model using a controller designed in an online fashion using a template model. To address this problem, this paper proposes a method to generate safety-preserving controllers for anchor models by performing reachability analysis on template models while bounding the modeling error. This paper describes how this reachable set can be incorporated into a Model Predictive Control framework to select controllers that result in safe walking on the anchor model in an online fashion. The method is illustrated on a 5-link RABBIT model, and is shown to allow the robot to walk safely while utilizing controllers designed in an online fashion.

Alexandra Carlson, Ram Vasudevan, Matthew Johnson-Roberson

Illumination effects in images, specifically cast shadows and shading, have been shown to decrease the performance of deep neural networks on a large number of vision-based detection, recognition and segmentation tasks in urban driving scenes. A key factor that contributes to this performance gap is the lack of `time-of-day' diversity within real, labeled datasets. There have been impressive advances in the realm of image to image translation in transferring previously unseen visual effects into a dataset, specifically in day to night translation. However, it is not easy to constrain what visual effects, let alone illumination effects, are transferred from one dataset to another during the training process. To address this problem, we propose deep learning framework, called Shadow Transfer, that can relight complex outdoor scenes by transferring realistic shadow, shading, and other lighting effects onto a single image. The novelty of the proposed framework is that it is both self-supervised, and is designed to operate on sensor and label information that is easily available in autonomous vehicle datasets. We show the effectiveness of this method on both synthetic and real datasets, and we provide experiments that demonstrate that the proposed method produces images of higher visual quality than state of the art image to image translation methods.

Ming-Yuan Yu, Ram Vasudevan, Matthew Johnson-Roberson

Knowing and predicting dangerous factors within a scene are two key components during autonomous driving, especially in a crowded urban environment. To navigate safely in environments, risk assessment is needed to quantify and associate the risk of taking a certain action. Risk assessment and planning is usually done by first tracking and predicting trajectories of other agents, such as vehicles and pedestrians, and then choosing an action to avoid collision in the future. However, few existing risk assessment algorithms handle occlusion and other sensory limitations effectively. This paper explores the possibility of efficient risk assessment under occlusion via both forward and backward reachability. The proposed algorithm can not only identify where the risk-induced factors are, but also be used for motion planning by executing low-level commands, such as throttle. The proposed method is evaluated on various four-way highly occluded intersections with up to five other vehicles in the scene. Compared with other risk assessment algorithms, the proposed method shows better efficiency, meaning that the ego vehicle reaches the goal at a higher speed. In addition, it also lowers the median collision rate by 7.5x.

Cyrus Anderson, Ram Vasudevan, Matthew Johnson-Roberson

Highway driving places significant demands on human drivers and autonomous vehicles (AVs) alike due to high speeds and the complex interactions in dense traffic. Merging onto the highway poses additional challenges by limiting the amount of time available for decision-making. Predicting others' trajectories accurately and quickly is crucial to safely executing maneuvers. Many existing prediction methods based on neural networks have focused on modeling interactions to achieve better accuracy while assuming the existence of observation windows over 3s long. This paper proposes a novel probabilistic model for trajectory prediction that performs competitively with as little as 400ms of observations. The proposed model extends a deterministic car-following model to the probabilistic setting by treating model parameters as unknown random variables and introducing regularization terms. A realtime inference procedure is derived to estimate the parameters from observations in this new model. Experiments on dense traffic in the NGSIM dataset demonstrate that the proposed method achieves state-of-the-art performance with both highly constrained and more traditional observation windows.

Junming Zhang, Manikandasriram Srinivasan Ramanagopal, Ram Vasudevan et al.

An accurate depth map of the environment is critical to the safe operation of autonomous robots and vehicles. Currently, either light detection and ranging (LIDAR) or stereo matching algorithms are used to acquire such depth information. However, a high-resolution LIDAR is expensive and produces sparse depth map at large range; stereo matching algorithms are able to generate denser depth maps but are typically less accurate than LIDAR at long range. This paper combines these approaches together to generate high-quality dense depth maps. Unlike previous approaches that are trained using ground-truth labels, the proposed model adopts a self-supervised training process. Experiments show that the proposed method is able to generate high-quality dense depth maps and performs robustly even with low-resolution inputs. This shows the potential to reduce the cost by using LIDARs with lower resolution in concert with stereo systems while maintaining high resolution.

Sean Vaskov, Shreyas Kousik, Hannah Larson et al.

As autonomous robots increasingly become part of daily life, they will often encounter dynamic environments while only having limited information about their surroundings. Unfortunately, due to the possible presence of malicious dynamic actors, it is infeasible to develop an algorithm that can guarantee collision-free operation. Instead, one can attempt to design a control technique that guarantees the robot is not-at-fault in any collision. In the literature, making such guarantees in real time has been restricted to static environments or specific dynamic models. To ensure not-at-fault behavior, a robot must first correctly sense and predict the world around it within some sufficiently large sensor horizon (the prediction problem), then correctly control relative to the predictions (the control problem). This paper addresses the control problem by proposing Reachability-based Trajectory Design for Dynamic environments (RTD-D), which guarantees that a robot with an arbitrary nonlinear dynamic model correctly responds to predictions in arbitrary dynamic environments. RTD-D first computes a Forward Reachable Set (FRS) offline of the robot tracking parameterized desired trajectories that include fail-safe maneuvers. Then, for online receding-horizon planning, the method provides a way to discretize predictions of an arbitrary dynamic environment to enable real-time collision checking. The FRS is used to map these discretized predictions to trajectories that the robot can track while provably not-at-fault. One such trajectory is chosen at each iteration, or the robot executes the fail-safe maneuver from its previous trajectory which is guaranteed to be not at fault. RTD-D is shown to produce not-at-fault behavior over thousands of simulations and several real-world hardware demonstrations on two robots: a Segway, and a small electric vehicle.

Corina Barbalata, Ram Vasudevan, Matthew Johnson-Roberson

This paper presents an interconnected control-planning strategy for redundant manipulators, subject to system and environmental constraints. The method incorporates low-level control characteristics and high-level planning components into a robust strategy for manipulators acting in complex environments, subject to joint limits. This strategy is formulated using an adaptive control rule, the estimated dynamic model of the robotic system and the nullspace of the linearized constraints. A path is generated that takes into account the capabilities of the platform. The proposed method is computationally efficient, enabling its implementation on a real multi-body robotic system. Through experimental results with a 7 DOF manipulator, we demonstrate the performance of the method in real-world scenarios.

Shreyas Kousik, Sean Vaskov, Fan Bu et al.

To operate with limited sensor horizons in unpredictable environments, autonomous robots use a receding-horizon strategy to plan trajectories, wherein they execute a short plan while creating the next plan. However, creating safe, dynamically-feasible trajectories in real time is challenging; and, planners must ensure persistent feasibility, meaning a new trajectory is always available before the previous one has finished executing. Existing approaches make a tradeoff between model complexity and planning speed, which can require sacrificing guarantees of safety and dynamic feasibility. This work presents the Reachability-based Trajectory Design (RTD) method for trajectory planning. RTD begins with an offline Forward Reachable Set (FRS) computation of a robot's motion when tracking parameterized trajectories; the FRS provably bounds tracking error. At runtime, the FRS is used to map obstacles to parameterized trajectories, allowing RTD to select a safe trajectory at every planning iteration. RTD prescribes an obstacle representation to ensure that obstacle constraints can be created and evaluated in real time while maintaining safety. Persistent feasibility is achieved by prescribing a minimum sensor horizon and a minimum duration for the planned trajectories. A system decomposition approach is used to improve the tractability of computing the FRS, allowing RTD to create more complex plans at runtime. RTD is compared in simulation with Rapidly-Exploring Random Trees and Nonlinear Model-Predictive Control. RTD is also demonstrated in randomly-crafted environments on two hardware platforms: a differential-drive Segway, and a car-like Rover. The proposed method is safe and persistently feasible across thousands of simulations and dozens of real-world hardware demos.

Alexandra Carlson, Katherine A. Skinner, Ram Vasudevan et al.

Performance on benchmark datasets has drastically improved with advances in deep learning. Still, cross-dataset generalization performance remains relatively low due to the domain shift that can occur between two different datasets. This domain shift is especially exaggerated between synthetic and real datasets. Significant research has been done to reduce this gap, specifically via modeling variation in the spatial layout of a scene, such as occlusions, and scene environmental factors, such as time of day and weather effects. However, few works have addressed modeling the variation in the sensor domain as a means of reducing the synthetic to real domain gap. The camera or sensor used to capture a dataset introduces artifacts into the image data that are unique to the sensor model, suggesting that sensor effects may also contribute to domain shift. To address this, we propose a learned augmentation network composed of physically-based augmentation functions. Our proposed augmentation pipeline transfers specific effects of the sensor model -- chromatic aberration, blur, exposure, noise, and color temperature -- from a real dataset to a synthetic dataset. We provide experiments that demonstrate that augmenting synthetic training datasets with the proposed learned augmentation framework reduces the domain gap between synthetic and real domains for object detection in urban driving scenes.

Ming-Yuan Yu, Ram Vasudevan, Matthew Johnson-Roberson

Navigating safely in urban environments remains a challenging problem for autonomous vehicles. Occlusion and limited sensor range can pose significant challenges to safely navigate among pedestrians and other vehicles in the environment. Enabling vehicles to quantify the risk posed by unseen regions allows them to anticipate future possibilities, resulting in increased safety and ride comfort. This paper proposes an algorithm that takes advantage of the known road layouts to forecast, quantify, and aggregate risk associated with occlusions and limited sensor range. This allows us to make predictions of risk induced by unobserved vehicles even in heavily occluded urban environments. The risk can then be used either by a low-level planning algorithm to generate better trajectories, or by a high-level one to plan a better route. The proposed algorithm is evaluated on intersection layouts from real-world map data with up to five other vehicles in the scene, and verified to reduce collision rates by 4.8x comparing to a baseline method while improving driving comfort.

Wonhui Kim, Manikandasriram Srinivasan Ramanagopal, Charles Barto et al.

This paper presents a novel dataset titled PedX, a large-scale multimodal collection of pedestrians at complex urban intersections. PedX consists of more than 5,000 pairs of high-resolution (12MP) stereo images and LiDAR data along with providing 2D and 3D labels of pedestrians. We also present a novel 3D model fitting algorithm for automatic 3D labeling harnessing constraints across different modalities and novel shape and temporal priors. All annotated 3D pedestrians are localized into the real-world metric space, and the generated 3D models are validated using a mocap system configured in a controlled outdoor environment to simulate pedestrians in urban intersections. We also show that the manual 2D labels can be replaced by state-of-the-art automated labeling approaches, thereby facilitating automatic generation of large scale datasets.

Matthew Porter, Arnav Joshi, Pedro Hespanhol et al.

A variety of anomaly detection schemes have been proposed to detect malicious attacks to Cyber-Physical Systems. Among these schemes, Dynamic Watermarking methods have been proven highly effective at detecting a wide range of attacks. Unfortunately, in contrast to other anomaly detectors, no method has been presented to design a Dynamic Watermarking detector to achieve a user-specified false alarm rate, or subsequently evaluate the capabilities of an attacker under such a selection. This paper describes methods to measure the capability of an attacker, to numerically approximate this metric, and to design a Dynamic Watermarking detector that can achieve a user-specified rate of false alarms. The performance of the Dynamic Watermarking detector is compared to three classical anomaly detectors in simulation and on a real-world platform. These experiments illustrate that the attack capability under the Dynamic Watermarking detector is comparable to those of classic anomaly detectors. Importantly, these experiments also make clear that the Dynamic Watermarking detector is consistently able to detect attacks that the other class of detectors are unable to identify.

Niankai Yang, Mohammad Reza Amini, Matthew Johnson-Roberson et al.

Improving endurance is crucial for extending the spatial and temporal operation range of autonomous underwater vehicles (AUVs). Considering the hardware constraints and the performance requirements, an intelligent energy management system is required to extend the operation range of AUVs. This paper presents a novel model predictive control (MPC) framework for energy-optimal point-to-point motion control of an AUV. In this scheme, the energy management problem of an AUV is reformulated as a surge motion optimization problem in two stages. First, a system-level energy minimization problem is solved by managing the trade-off between the energies required for overcoming the positive buoyancy and surge drag force in static optimization. Next, an MPC with a special cost function formulation is proposed to deal with transients and system dynamics. A switching logic for handling the transition between the static and dynamic stages is incorporated to reduce the computational efforts. Simulation results show that the proposed method is able to achieve near-optimal energy consumption with considerable lower computational complexity.

Alexandra Carlson, Katherine A. Skinner, Ram Vasudevan et al.

Recent work has focused on generating synthetic imagery to increase the size and variability of training data for learning visual tasks in urban scenes. This includes increasing the occurrence of occlusions or varying environmental and weather effects. However, few have addressed modeling variation in the sensor domain. Sensor effects can degrade real images, limiting generalizability of network performance on visual tasks trained on synthetic data and tested in real environments. This paper proposes an efficient, automatic, physically-based augmentation pipeline to vary sensor effects --chromatic aberration, blur, exposure, noise, and color cast-- for synthetic imagery. In particular, this paper illustrates that augmenting synthetic training datasets with the proposed pipeline reduces the domain gap between synthetic and real domains for the task of object detection in urban driving scenes.

Henry O. Jacobs, Owen K. Hughes, Matthew Johnson-Roberson et al.

The success of autonomous systems will depend upon their ability to safely navigate human-centric environments. This motivates the need for a real-time, probabilistic forecasting algorithm for pedestrians, cyclists, and other agents since these predictions will form a necessary step in assessing the risk of any action. This paper presents a novel approach to probabilistic forecasting for pedestrians based on weighted sums of ordinary differential equations that are learned from historical trajectory information within a fixed scene. The resulting algorithm is embarrassingly parallel and is able to work at real-time speeds using a naive Python implementation. The quality of predicted locations of agents generated by the proposed algorithm is validated on a variety of examples and considerably higher than existing state of the art approaches over long time horizons.

Matthew Johnson-Roberson, Charles Barto, Rounak Mehta et al.

Deep learning has rapidly transformed the state of the art algorithms used to address a variety of problems in computer vision and robotics. These breakthroughs have relied upon massive amounts of human annotated training data. This time consuming process has begun impeding the progress of these deep learning efforts. This paper describes a method to incorporate photo-realistic computer images from a simulation engine to rapidly generate annotated data that can be used for the training of machine learning algorithms. We demonstrate that a state of the art architecture, which is trained only using these synthetic annotations, performs better than the identical architecture trained on human annotated real-world data, when tested on the KITTI data set for vehicle detection. By training machine learning algorithms on a rich virtual world, real objects in real scenes can be learned and classified using synthetic data. This approach offers the possibility of accelerating deep learning's application to sensor-based classification problems like those that appear in self-driving cars. The source code and data to train and validate the networks described in this paper are made available for researchers.