ICRA 2023 Papers

We present an optimization-based framework for rearranging indoor furniture to accommodate human-robot co-activities better. The rearrangement aims to afford sufficient accessible space for robot activities without compromising everyday human activities. To retain human activities, our algorithm preserves the functional relations among furniture by integrating spatial and semantic co-occurrence extracted from SUNCG and ConceptNet, respectively. By defining the robot's accessible space by the amount of open space it can traverse and the number of objects it can reach, we formulate the rearrangement for human-robot co-activity as an optimization problem, solved by adaptive simulated annealing (ASA) and covariance matrix adaptation evolution strategy (CMA-ES). Our experiments on the SUNCG dataset quantitatively show that rearranged scenes provide a robot with 14% more accessible space and 30% more objects to interact with on average. The quality of the rearranged scenes is qualitatively validated by a human study, indicating the efficacy of the proposed strategy.

Event cameras are becoming increasingly popular in robotics and computer vision due to their beneficial properties, e.g., high temporal resolution, high bandwidth, almost no motion blur, and low power consumption. However, these cameras remain expensive and scarce in the market, making them inaccessible to the majority. Using event simulators minimizes the need for real event cameras to develop novel algorithms. However, due to the computational complexity of the simulation, the event streams of existing simulators cannot be generated in real-time but rather have to be pre-calculated from existing video sequences or pre-rendered and then simulated from a virtual 3D scene. Although these offline generated event streams can be used as training data for learning tasks, all response time dependent applications cannot benefit from these simulators yet, as they still require an actual event camera. This work proposes simulation methods that improve the performance of event simulation by two orders of magnitude (making them real-time capable) while remaining competitive in the quality assessment.

Background subtraction is an important topic in computer vision and video analysis. It is challenging to robustly segment foreground and background in complex scenarios. In the literature there are efforts to address some of the main challenges such as illumination change, dynamic backgrounds, hard shadows, and intermittent object motion. However, most of the research has focused on applying advanced mathematical and machine learning models rather than on improving performance in real-time applications. In this paper, we devise a method named EGMM to efficiently handle the illumination change problem and also operate at a real-time execution speed on commodity PC hardware. EGMM is an ensemble algorithm that fuses multiple Gaussian Mixture Models operating on gradient, texture and color features. Detection and removal of shadows is done using a chromaticity-based approach, and spatio-temporal history of foreground blobs is used to handle intermittent object motion. We benchmarked EGMM by creating datasets for two light change scenarios. The results demonstrate that EGMM achieves robust performance in complex illumination change cases, outperforms some state-of-the-art algorithms, and runs at 100 fps (GPU) at 1280\times7201280\times720 resolution. Moreover, experiments using the 2012 CDnet dataset show that EGMM achieves generally good performance in varying scenes with overall results better than conventional methods and runs at 1000 fps (GPU) at 320\times 240320\times 240 resolution.

Phase-only acoustic holography is a fundamental and promising technique for contactless robotic manipulation. Through independently controlling phase-only hologram (POH) of phase array of transducers (PAT) and simultaneously driving each channel by sophisticated circuits, a certain acoustic field is dynamically generated in working medium (e.g., air, water or biological tissues) at certain moment. The phase profile of PAT is required dynamically and precisely as per arbitrary expected acoustic field for the sake of versatile and stable robotic manipulation. However, the most conventional methods rely on iterative optimization algorithms which are inevitably time-consuming and probably non-convergent, moreover hindering versatility and fidelity of acoustic robotic manipulation. To address these issues, this paper reports a real-time phase-only acoustic holography algorithm by virtue of iterative unsupervised learning. Using a physics model to construct two queues, which we refer to as experience pools, data pairs consisting of a target acoustic amplitude hologram in expected acoustic field and corresponding POH of PAT are collected on-the-fly, circumventing costly preparation of annotated dataset in advance. With iterative learning between neural network training and experience pools update, both the solution of objective inverse mapping and the adaptation for arbitrary desired acoustic field are mutually enhanced. The experiments and results validated that the proposed approach surpasses previous algorithms in terms of real time and precision.

Trajectory planning and control have historically been separated into two modules in automated driving stacks. Trajectory planning focuses on higher-level tasks like avoiding obstacles and staying on the road surface, whereas the controller tries its best to follow an ever changing reference trajectory. We argue that this separation is (1) flawed due to the mismatch between planned trajectories and what the controller can feasibly execute, and (2) unnecessary due to the flexibility of the model predictive control (MPC) paradigm. Instead, in this paper, we present a unified MPC-based trajectory planning and control scheme that guarantees feasibility with respect to road boundaries, the static and dynamic environment, and enforces passenger comfort constraints. The scheme is evaluated rigorously in a variety of scenarios focused on proving the effectiveness of the optimal control problem (OCP) design and real-time solution methods. The prototype code will be released at github.com/WATonomous/control.

We propose a novel real-time LiDAR intensity image-based simultaneous localization and mapping method, which addresses the geometry degeneracy problem in un-structured environments. Traditional LiDAR-based front-end odometry mostly relies on geometric features such as points, lines and planes. A lack of these features in the environment can lead to the failure of the entire odometry system. To avoid this problem, we extract feature points from the LiDAR-generated point cloud that match features identified in LiDAR intensity images. We then use the extracted feature points to perform scan registration and estimate the robot ego-movement. For the back-end, we jointly optimize the distance between the corresponding feature points, and the point to plane distance for planes identified in the map. In addition, we use the features extracted from intensity images to detect loop closure candidates from previous scans and perform pose graph optimization. Our experiments show that our method can run in real time with high accuracy and works well with illumination changes, low-texture, and unstructured environments.

Real-time learning is crucial for robotic agents adapting to ever-changing, non-stationary environments. A common setup for a robotic agent is to have two different computers simultaneously: a resource-limited local computer tethered to the robot and a powerful remote computer connected wirelessly. Given such a setup, it is unclear to what extent the performance of a learning system can be affected by resource limitations and how to efficiently use the wirelessly connected powerful computer to compensate for any performance loss. In this paper, we implement a real-time learning system called the Remote-Local Distributed (ReLoD) system to distribute computations of two deep reinforcement learning (RL) algorithms, Soft Actor-Critic (SAC) and Proximal Policy Optimization (PPO), between a local and a remote computer. The performance of the system is evaluated on two vision-based control tasks developed using a robotic arm and a mobile robot. Our results show that SAC's performance degrades heavily on a resource-limited local computer. Strikingly, when all computations of the learning system are deployed on a remote workstation, SAC fails to compensate for the performance loss, indicating that, without careful consideration, using a powerful remote computer may not result in performance improvement. However, a carefully chosen distribution of computations of SAC consistently and substantially improves its performance on both tasks. On the other hand, the performance of PPO remains largely unaffected by the distribution of computations. In addition, when all computations happen solely on a powerful tethered computer, the performance of our system remains on par with an existing system that is well-tuned for using a single machine. ReLoD is the only publicly available system for real-time RL that applies to multiple robots for vision-based tasks. The source code can be found at https://github.com/rlai-lab/relod

Vessel transit in ice-covered waters poses unique challenges in safe and efficient motion planning. When the concentration of ice is high, it may not be possible to find collision-free trajectories. Instead, ice can be pushed out of the way if it is small or if contact occurs near the edge of the ice. In this work, we propose a real-time navigation framework that minimizes collisions with ice and distance travelled by the vessel. We exploit a lattice-based planner with a cost that captures the ship interaction with ice. To address the dynamic nature of the environment, we plan motion in a receding horizon manner based on updated vessel and ice state information. Further, we present a novel planning heuristic for evaluating the cost-to-go, which is applicable to navigation in a channel without a fixed goal location. The performance of our planner is evaluated across several levels of ice concentration both in simulated and in real-world experiments.

This paper proposes a real-time model predictive control (MPC) strategy for accomplishing multiple tasks using robots within a finite-time horizon. In industrial robotic applications, it is crucial to consider various constraints to ensure that joint position, velocity, and torque limits are not exceeded. In addition, singularity-free and smooth motions require executing tasks continuously and safely. Instead of formulating nonlinear MPC problems, we devise linear MPC problems using kinematic and dynamic models linearized along nominal trajectories produced by hierarchical controllers. These linear MPC problems are solvable via the use of Quadratic Pro-gramming; therefore, we significantly reduce the computation time of the proposed MPC framework so the resulting update frequency is higher than 1 kHz. Our proposed MPC framework is more efficient in reducing task tracking errors than a baseline based on operational space control (OSC). We validate our approach in numerical simulations and in real experiments using an industrial manipulator. More specifically, we deploy our method in two practical scenarios for robotic logistics: 1) controlling a robot carrying heavy payloads while accounting for torque limits, and 2) controlling the end-effector while avoiding singularities.

Robots performing mobile manipulation in unstructured environments must identify grasp affordances quickly and with robustness to perception noise. Yet in domains such as underwater manipulation, where perception noise is severe, computation is constrained, and the environment is dynamic, existing techniques fail. They are too computationally demanding, or too sensitive to noise to allow for closed loop grasping or dynamic replanning, or do not consider 6-DOF grasps. We present a novel grasp synthesis network, TSGrasp, that uses spatio-temporal sparse convolution to process a streaming point cloud in real time. The network generates 6-DOF grasps at greater speed and with less memory than Contact GraspNet, a state-of-the-art algorithm based on Point-Net++. By considering information from multiple successive frames of depth video, TSGrasp boosts robustness to noise or temporary self-occlusion and allows more grasps to be rapidly identified. Our grasp synthesis system was successfully demonstrated in an underwater environment with a Blueprint Labs Bravo robotic arm.

This paper proposes the Real-Time Fast Marching Tree (RT-FMT), a real-time planning algorithm that features local and global path generation, multiple-query planning, and dynamic obstacle avoidance. During the search, RT-FMT quickly looks for the global solution and, in the meantime, generates local paths that can be used by the robot to start execution faster. In addition, our algorithm constantly rewires the tree to keep branches from forming inside the dynamic obstacles and to maintain the tree root near the robot, which allows the tree to be reused multiple times for different goals. Our algorithm is based on the planners Fast Marching Tree (FMT*) and Real-time Rapidly-Exploring Random Tree (RT-RRT*). We show via simulations that RT-FMT outperforms RT- RRT* in both execution cost and arrival time, in most cases. Moreover, we also demonstrate via simulation that it is worthwhile taking the local path before the global path is available in order to reduce arrival time, even though there is a small possibility of taking an inferior path.

Gait parameters, including walking speed and stride length, are crucial indicators of health status and rehabilitation progress for individuals using wearable robots for exercise or rehabilitation. These metrics play a crucial role in monitoring progress and adjusting training programs, thereby fostering greater engagement in the training. In this paper, we present methods for estimating walking speed and stride length using sensors in wearable hip exoskeleton GEMS-H. Our study collected data from 79 middle-aged healthy individuals walking on a treadmill while wearing GEMS-H under various assistance conditions. To estimate walking speed, we evaluated linear regression models, deep neural networks, and ensemble models using different combinations of joint encoders and an IMU in the GEMS-H hip exoskeleton to form various sets of features. The ensemble of deep neural networks using only 6-DOF IMU signals as features achieved the lowest root-mean-square error (RMSE) for walking speed estimation, which was 0.066 m/s. We also present an algorithm for real-time stride length estimation, building on one of the speed estimation models. The speed and stride length estimation model was tested on 12 middle-aged healthy subjects walking in GEMS-H overground, yielding an RMSE of 0.060 m/s for speed and 7.1 cm for stride length.

This paper addresses real-time dense 3D reconstruction for a resource-constrained Autonomous Underwater Vehicle (AUV). Underwater vision-guided operations are among the most challenging as they combine 3D motion in the presence of external forces, limited visibility, and absence of global positioning. Obstacle avoidance and effective path planning require online dense reconstructions of the environment. Autonomous operation is central to environmental monitoring, marine archaeology, resource utilization, and underwater cave exploration. To address this problem, we propose to use SVIn2, a robust VIO method, together with a real-time 3D reconstruction pipeline. We provide extensive evaluation on four challenging underwater datasets. Our pipeline produces comparable reconstruction with that of COLMAP, the state-of-the-art offline 3D reconstruction method, at high frame rates on a single CPU.

We present a lightweight, decentralized algorithm for navigating multiple nonholonomic agents through challenging environments with narrow passages. Our key idea is to allow agents to yield to each other in large open areas instead of narrow passages, to increase the success rate of conventional decentralized algorithms. At pre-processing time, our method computes a medial axis for the freespace. A reference trajectory is then computed and projected onto the medial axis for each agent. During run time, when an agent senses other agents moving in the opposite direction, our algorithm uses the medial axis to estimate a Point of Impact (POI) as well as the available area around the POI. If the area around the POI is not large enough for yielding behaviors to be successful, we shift the POI to nearby large areas by modulating the agent's reference trajectory and traveling speed. We evaluate our method on a row of 4 environments with up to 15 robots, and we find our method incurs a marginal computational overhead of 10–30 ms on average, achieving real-time performance. Afterward, our planned reference trajectories can be tracked using local navigation algorithms to achieve up to a 100% higher success rate over local navigation algorithms alone.

Autonomous suturing has been a long-sought-after goal for surgical robotics. Outside of staged environments, accurate localization of suture needles is a critical foundation for automating various suture needle manipulation tasks in the real world. When localizing a needle held by a gripper, previous work usually tracks them separately without considering their relationship. Because of the significant errors that can arise in the stereo-triangulation of objects and instruments, their reconstructions may often not be consistent. This can lead to unrealistic tool-needle grasp reconstructions that are infeasible. Instead, an obvious strategy to improve localization would be to leverage constraints that arise from contact, thereby constraining reconstructions of objects and instruments into a jointly feasible space. In this work, we consider feasible grasping constraints when tracking the 6D pose of an in-hand suture needle. We propose a reparameterization trick to define a new state space for describing a needle pose, where grasp constraints can be easily defined and satisfied. Our proposed state space and feasible grasping constraints are then incorporated into Bayesian filters for real-time needle localization. In the experiments, we show that our constrained methods outperform previous unconstrained tracking approaches and demonstrate the importance of incorporating feasible grasping constraints into automating suture needle manipulation tasks.

Offline reinforcement learning (ORL) holds great promise for robot learning due to its ability to learn from arbitrary pre-generated experience. However, current ORL benchmarks are almost entirely in simulation and utilize contrived datasets like replay buffers of online RL agents or sub-optimal trajectories, and thus hold limited relevance for real-world robotics. In this work (Real-ORL), we posit that data collected from safe operations of closely related tasks are more practical data sources for real-world robot learning. Under these settings, we perform an extensive (6500+ trajectories collected over 800+ robot hours and 270+ human labor hour) empirical study evaluating generalization and transfer capabilities of representative ORL methods on four real-world tabletop manipulation tasks. Our study finds that ORL and imitation learning prefer different action spaces, and that ORL algorithms can generalize from leveraging offline heterogeneous data sources and outperform imitation learning. We release our dataset and implementations at URL: https://sites.google.com/view/real-orl.

We propose a novel Branch-and-Bound method for reachability analysis of neural networks in both open-loop and closed-loop settings. Our idea is to first compute accurate bounds on the Lipschitz constant of the neural network in certain directions of interest offline using a convex program. We then use these bounds to obtain an instantaneous but conservative polyhedral approximation of the reachable set using Lipschitz continuity arguments. To reduce conservatism, we incorporate our bounding algorithm within a branching strategy to decrease the over-approximation error within an arbitrary accuracy. We then extend our method to reachability analysis of control systems with neural network controllers. Finally, to capture the shape of the reachable sets as accurately as possible, we use sample trajectories to inform the directions of the reachable set over-approximations using Principal Com-ponent Analysis (PCA). We evaluate the performance of the proposed method in several open-loop and closed-loop settings.

Generating feasible robot motions in real-time requires achieving multiple tasks (i.e., kinematic requirements) simultaneously. These tasks can have a specific goal, a range of equally valid goals, or a range of acceptable goals with a preference toward a specific goal. To satisfy multiple and potentially competing tasks simultaneously, it is important to exploit the flexibility afforded by tasks with a range of goals. In this paper, we propose a real-time motion generation method that accommodates all three categories of tasks within a single, unified framework and leverages the flexibility of tasks with a range of goals to accommodate other tasks. Our method incorporates tasks in a weighted-sum multiple-objective optimization structure and uses barrier methods with novel loss functions to encode the valid range of a task. We demonstrate the effectiveness of our method through a simulation experiment that compares it to state-of-the-art alternative approaches, and by demonstrating it on a physical camera-in-hand robot that shows that our method enables the robot to achieve smooth and feasible camera motions.

The awareness about moving objects in the surroundings of a self-driving vehicle is essential for safe and reliable autonomous navigation. The interpretation of LiDAR and camera data achieves exceptional results but typically requires to accumulate and process temporal sequences of data in order to extract motion information. In contrast, radar sensors, which are already installed in most recent vehicles, can overcome this limitation as they directly provide the Doppler velocity of the detections and, hence incorporate instantaneous motion information within a single measurement. In this paper, we tackle the problem of moving object segmentation in noisy radar point clouds. We also consider differentiating parked from moving cars, to enhance scene understanding. Instead of exploiting temporal dependencies to identify moving objects, we develop a novel transformer-based approach to perform single-scan moving object segmentation in sparse radar scans accurately. The key to our Radar Velocity Transformer is to incorporate the valuable velocity information throughout each module of the network, thereby enabling the precise segmentation of moving and non-moving objects. Additionally, we propose a transformer-based upsampling, which enhances the performance by adaptively combining information and over-coming the limitation of interpolation of sparse point clouds. Finally, we create a new radar moving object segmentation benchmark based on the RadarScenes dataset and compare our approach to other state-of-the-art methods. Our network runs faster than the frame rate of the sensor and shows superior segmentation results using only single-scan radar data.