ICRA 2022 论文列表

Deep reinforcement learning (RL) has shown promising results in the motion planning of manipulators. However, no method guarantees the safety of highly dynamic obstacles, such as humans, in RL-based manipulator control. This lack of formal safety assurances prevents the application of RL for manipulators in real-world human environments. Therefore, we propose a shielding mechanism that ensures ISO- verified human safety while training and deploying RL algorithms on manipulators. We utilize a fast reachability analysis of humans and manipulators to guarantee that the manipulator comes to a complete stop before a human is within its range. Our proposed method guarantees safety and significantly improves the RL performance by preventing episode-ending collisions. We demonstrate the performance of our proposed method in simulation using human motion capture data.

LiDAR point cloud panoptic segmentation, including both semantic and instance segmentation, plays a critical role in meticulous scene understanding for autonomous driving. Existing 3D voxelized approaches either utilize 3D sparse convolution that only focuses on local scene understanding, or add extra and time-consuming PointNet branch to capture global feature structures. To address these limitations, we propose an end-to-end Prototype-Voxel Contrastive Learning (PVCL) framework for learning stable and discriminative semantic representations, which includes voxel-level and prototype-level contrastive learning (CL). The voxel-level CL decreases intra-class distance and increases inter-class distance among sample representations, while the prototype-level CL further reduces the dependence of CL on negative sampling and avoids the influence of outliers from the same class, enabling PVCL to be more effective for outdoor point cloud panoptic segmentation. Extensive experiments are conducted on the public point cloud panoptic segmentation datasets, Semantic-KITTI and nuScenes, where evaluations and ablation studies demonstrate PVCL achieves superior performance compared with the state-of-the-art. Our approach ranks the top on the public leaderboard of Semantic-KITTI at the time of submission, and surpasses the published 2nd rank, EfficientLPS, by 1.7% in PQ.

Uncertainty pervades through the modern robotic autonomy stack, with nearly every component (e.g., sensors, detection, classification, tracking, behavior prediction) producing continuous or discrete probabilistic distributions. Trajectory forecasting, in particular, is surrounded by uncertainty as its inputs are produced by (noisy) upstream perception and its outputs are predictions that are often probabilistic for use in downstream planning. However, most trajectory forecasting methods do not account for upstream uncertainty, instead taking only the most-likely values. As a result, perceptual uncer-tainties are not propagated through forecasting and predictions are frequently overconfident. To address this, we present a novel method for incorporating perceptual state uncertainty in trajectory forecasting, a key component of which is a new statistical distance-based loss function which encourages predicting uncertainties that better match upstream perception. We evaluate our approach both in illustrative simulations and on large-scale, real-world data, demonstrating its efficacy in propagating perceptual state uncertainty through prediction and producing more calibrated predictions.

Quality Diversity (QD) algorithms in population-based reinforcement learning aim to optimize agents' returns and diversity among the population simultaneously. It is conducive to solving exploration problems in reinforcement learning and potentially getting multiple good and diverse strategies. However, previous methods typically define behavioral embedding in action space or outcome space, which neglect trajectory characteristics during the execution process. In this paper, we introduce a trajectory embedding model trained by Variational Autoencoder with similarity constraint to characterize trajectory features. Based on that, we propose a hierarchical trajectory-space exploration (HTSE) framework using Determinantal Point Processes (DPP) to generate high-quality and diverse solutions in the selection and mutation process. The experimental results show that our HTSE method effectively completes several simulated tasks, outperforming other Quality-Diversity Reinforcement Learning algorithms.

Reproducing real world dynamics in simulation is critical for the development of new control and perception methods. This task typically involves the estimation of simu-lation parameter distributions from observed rollouts through an inverse inference problem characterized by multi-modality and skewed distributions. We address this challenging problem through a novel Bayesian inference approach that approximates a posterior distribution over simulation parameters given real sensor measurements. By extending the commonly used Gaus-sian likelihood model for trajectories via the multiple-shooting formulation, our gradient-based particle inference algorithm, Stein Variational Gradient Descent, is able to identify highly nonlinear, underactuated systems. We leverage GPU code gen-eration and differentiable simulation to evaluate the likelihood and its gradient for many particles in parallel. Our algorithm infers nonparametric distributions over simulation parame-ters more accurately than comparable baselines and handles constraints over parameters efficiently through gradient-based optimization. We evaluate estimation performance on several physical experiments. On an underactuated mechanism where a 7-DOF robot arm excites an object with an unknown mass configuration, we demonstrate how the inference technique can identify symmetries between the parameters and provide highly accurate predictions. Website: https://uscresl.github.io/prob-diff-sim

Navigation in close proximity with pedestrians is a challenge on the way to fully automated vehicles. Pedestrian-friendly navigation requires an understanding of pedestrian reaction and intention. Merely safety based reactive systems can lead to sub-optimal navigation solutions resulting in the freezing of the vehicle in many scenarios. Moreover, a strictly reactive method can produce unnatural driving patterns which cannot guarantee the legibility or social acceptance of the automated vehicle. This work presents a proactive maneuvering method adapted to navigation in close interaction with pedestrians using a dynamic channel approach. The method allows to proactively explore the navigation options based on anticipating pedestrians cooperation. The navigation is tested in frontal and lateral crossing scenarios with variable space density. The system is implemented under ROS, and compared with the probabilistic Risk-RRT planning method. The results are evaluated based on the safety and comfort of the pedestrians, and the quality of the vehicle's trajectory.

We present a novel path-planning algorithm to reduce localization error for a network of robots cooperatively localizing via inter-robot range measurements. The quality of localization with range measurements depends on the configuration of the network, and poor configurations can cause substantial localization errors. To reduce the effect of network configuration on localization error for moving networks we consider various optimality measures of the Fisher information matrix (FIM), which have well-known relationships with localization error. We pose a trajectory planning problem with constraints on the FIM optimality measures. By constraining these optimality measures we can control the statistical properties of the localization error. To efficiently generate trajectories which satisfy these FIM constraints we present a prioritized planner which leverages graph-based planning and properties of the range-only FIM. We demonstrate in simulations that the trajectories generated by our algorithm reduce worst-case localization error by up to 42% in comparison to existing planners and can scalably plan distance-efficient trajectories in complicated environments for large numbers of robots.

Origami/kirigami robotics are opening a path that leads to lightweight, compact, and expandable robots. However, it is generally challenging to design agile motions for origami/kirigami robots due to their size and the intrinsic limitation of the materials. In this paper, we propose to use the bistability of the waterbomb base structure to generate the swift motion of the robots. We evaluate the bistability of the waterbomb-based structure and build origami jumpers with different configurations of the body to help analyze the behavior of the waterbomb base bistable structure. The jumper is actuated by a phase change liquid pouch actuator. Our jumper is lightweight (0.3 g), flattenable, and able to jump to more than 12 times of its diameter and 112 times of its height.

Powered knee prosthesis commonly improves locomotion of above-knee amputees by generating net positive mechanical work at the knee joint which is especially required for movements with active knee extension and flexion such as sit-to-stand maneuvers, level-ground walking with various walking speed, stair/slope ascent ambulation and so forth. These studies tend to refer and trace normal human locomotion, and have amputees move “normally”, which is quite difficult for patients with severe disability such as four-limb deficiency. In this study, a powered knee prosthesis with small-scale, light-weight and affordable series elastic actuator (PKP-SEA) is developed and adopted to a walking rehabilitation of a patient with four-limb deficiency. A subject is 45 years old and has no experience to walk on prosthetic devices in his life. During the experiment, walking performance is investigated by turning PKP-SEA control parameters. As a result, the PKP-SEA shows capability of improving his walking performance in terms of walking speed and consistency not by replicating normal human walking, but suggesting an original walking gait based on his body condition and remaining functionality.

As technology advances, the need for safe, efficient, and collaborative human-robot-teams has become increasingly important. One of the most fundamental collaborative tasks in any setting is the object handover. Human-to-robot handovers can take either of two approaches: (1) direct hand-to-hand or (2) indirect hand-to-placement-to-pick-up. The latter approach ensures minimal contact between the human and robot but can also result in increased idle time due to having to wait for the object to first be placed down on a surface. To minimize such idle time, the robot must preemptively predict the human intent of where the object will be placed. Furthermore, for the robot to preemptively act in any sort of productive manner, predictions and motion planning must occur in real-time. We introduce a novel prediction-planning pipeline that allows the robot to preemptively move towards the human agent's intended placement location using gaze and gestures as model inputs. In this paper, we investigate the performance and drawbacks of our early intent predictor-planner as well as the practical benefits of using such a pipeline through a human-robot case study.

Predicting the future motion of traffic agents is crucial for safe and efficient autonomous driving. To this end, we present PredictionNet, a deep neural network (DNN) that predicts the motion of all surrounding traffic agents together with the ego-vehicle's motion. All predictions are probabilistic and are represented in a simple top-down rasterization that allows an arbitrary number of agents. Conditioned on a multi-layer map with lane information, the network outputs future positions, velocities, and backtrace vectors jointly for all agents including the ego-vehicle in a single pass. Trajectories are then extracted from the output. The network can be used to simulate realistic traffic, and it produces competitive results on popular benchmarks. More importantly, it has been used to successfully control a real-world vehicle for hundreds of kilometers, by combining it with a motion planning/control subsystem. The network runs faster than real-time on an embedded GPU, and the system shows good generalization (across sensory modalities and locations) due to the choice of input representation. Furthermore, we demonstrate that by extending the DNN with reinforcement learning (RL), it can better handle rare or unsafe events like aggressive maneuvers and crashes.

With the advancement in computing and robotics, it is necessary to develop fluent and intuitive methods for inter-acting with digital systems, augmented/virtual reality (AR/VR) interfaces, and physical robotic systems. Hand movement recognition is widely used to enable such interaction. Hand configuration classification and metacarpophalangeal (MCP) joint angle detection are important for a comprehensive reconstruction of hand motion. Surface electromyography (sEMG) and other technologies have been used for the detection of hand motions. Ultrasound images of the forearm offer a way to visualize the internal physiology of the hand from a musculoskeletal perspective. Recent works have shown that these images can be classified using machine learning to predict various hand configurations. In this paper, we propose a Convolutional Neu-ral Network (CNN) based deep learning pipeline for predicting the MCP joint angles. We supplement our results by using a Support Vector Classifier (SVC) to classify the ultrasound information into several predefined hand configurations based on activities of daily living (ADL). Ultrasound data from the forearm were obtained from six subjects who were instructed to move their hands according to predefined hand configurations relevant to ADLs. Motion capture data was acquired as the ground truth for hand movements at three speeds (0.5 Hz, 1 Hz, and 2 Hz) for the index, middle, ring, and pinky fingers. We demonstrated the perfect prediction of hand configurations through SVC classification and a correspondence between the predicted MCP joint angles and the actual MCP joint angles for the fingers, with an average root mean square error of 7.35 degrees. A low latency (6.25 – 9.10 Hz) pipeline was implemented for the prediction of both MCP joint angles and hand configuration estimation aimed for real-time implementation.

Depth images usually contain pixels with invalid measurements. This paper presents a deep learning approach that receives as input a partially-known volumetric model of the environment and a camera pose, and it predicts the probability that a pixel would contain a valid depth measurement if a camera was placed at the given pose. The proposed network architecture consists of a 3D Convolutional Neural Network (CNN) module and a 2D CNN module, connected by a deep learning attention-based projection module. The method was integrated into a CNN-based probabilistic Next Best View plan-ner, resulting in a more realistic prediction of the information gain for each possible viewpoint with respect to state of the art approaches. Experiments were carried out in tabletop scenarios using a robot manipulator with an eye-in-hand depth camera.

Parkinson's disease is a severe neurodegenerative disorder that affects sensorimotor control. In particular, several gait impairments are reported, including a decrease of long-range autocorrelations in stride duration time series. This complex statistics is potentially a biomarker of the risk of falling. This paper aims at developing model-based predictions about the loss of long-range autocorrelations in the gait of Parkinsonian patients, and how these autocorrelations can be restored by an oscillator-based walking assistance. Using a Super Central Pattern Generator model coupled with an adaptive oscillator, we show that this type of assistance has the potential to improve long-range autocorrelations in time series of Parkinsonian walkers. This requires however to tune the adaptive oscillator with slow learning gains, raising challenges for porting this method to an actual device.

Pilots operating aircraft in non-towered terminal airspace rely on their situational awareness and prior knowledge to predict the future trajectories of other agents. These predictions are conditioned on the past trajectories of other agents, agent-agent social interactions and environmental context such as airport location and weather. This paper provides a dataset, TrajAir, that captures this behaviour in non-towered terminal airspace around a regional airport. We also present a baseline socially-aware trajectory prediction algorithm, TrajAirNet, that uses the dataset to predict the trajectories of all agents. The dataset is collected for 111 days over 8 months and contains ADS-B transponder data along with the corresponding METAR weather data. The data is processed to be used as a benchmark with other publicly available social navigation datasets. To the best of the authors' knowledge, this is the first 3D social aerial navigation dataset, thus introducing social navigation for autonomous aviation. TrajAirNet combines state-of-the-art modules in social navigation to provide predictions in a static environment with a dynamic context. Both the TrajAir dataset and TrajAirNet prediction algorithm are open-source. [Dataset] 11Dataset: https://theairlab.org/trajair/ [Code]22Codebase: https://github.com/castacks/trajairnet [Video]33Video: https://youtu.be/e1AQXrxB2gw

Robotic tree pruning requires highly precise manipulator control in order to accurately align a cutting implement with the desired pruning point at the correct angle. Simultaneously, the robot must avoid applying excessive force to rigid parts of the environment such as trees, support posts, and wires. In this paper, we propose a hybrid control system that uses a learned vision-based controller to initially align the cutter with the desired pruning point, taking in images of the environment and outputting control actions. This controller is trained entirely in simulation, but transfers easily to real trees via a neural network which transforms raw images into a simplified, segmented representation. Once contact is established, the system hands over control to an interaction controller that guides the cutter pivot point to the branch while minimizing interaction forces. With this simple, yet novel, approach we demonstrate an improvement of over 30 percentage points in accuracy over a baseline controller that uses camera depth data.

Monitoring individual plants and computing precise 3D reconstructions is highly relevant for crop breeding. In the conventional breeding approach, humans measure phenotypic traits by hand, requiring substantial manual labor. This paper addresses precise 3D plant reconstructions in a crop field or breeding plot based on UAV imagery. We explicitly address the challenges resulting from the thin structures of leaves and naturally occurring self-occlusions. We combine photogrammetric bundle adjustment with a template-based matching approach and produce accurate 3D models that allow us to derive common, geometric traits used by breeders to phenotype plants. We provide a thorough experimental evaluation on commercially used sugar beet breeding plots to illustrate the capabilities of our method as well as its real world applicability.

A multi-fingered hand that can grasp and manipulate a variety of objects is an option for assisting people in their daily lives. However, the range of torque output that can be handled by the multi-fingered hand is very limited compared to the capability of the human hand. In this paper, we introduce a new multi-fingered hand consisting of a dynamic pulley and a linkage mechanism, aiming to achieve a human-like output torque with a human-like size. The proposed multi-fingered hand can achieve a fingertip force of 50N, which is equivalent to that of a human, and at the same time can perform delicate operations such as picking up a coin on a desk. In addition, we realized the stay-on-tab opening task of a can by utilizing fingertip strength.

As service robots begin to be deployed to assist humans, it is important for them to be able to perform a skill as ubiquitous as pouring. Specifically, we focus on the task of pouring an exact amount of water without any environmental instrumentation, that is, using only the robot's own sensors to perform this task in a general way robustly. In our approach we use a simple PID controller which uses the measured change in weight of the held container to supervise the pour. Unlike previous methods which use specialized force-torque sensors at the robot wrist, we use our robot joint torque sensors and investigate the added benefit of tactile sensors at the fingertips. We train three estimators from data which regress the poured weight out of the source container and show that we can accurately pour within 10 ml of the target on average while being robust enough to pour at novel locations and with different grasps on the source container.