IROS 2022 论文列表

In this work, we introduce a new perspective for learning transferable content in multi-task imitation learning. Humans are capable of transferring skills and knowledge. If we can cycle to work and drive to the store, we can also cycle to the store and drive to work. We take inspiration from this and hypothesize the latent memory of a policy network can be disentangled into two partitions. These contain either the knowledge of the environmental context for the task or the generalisable skill needed to solve the task. This allows an improved training efficiency and better generalization over previously unseen combinations of skills in the same environment, and the same task in unseen environments. We used the proposed approach to train a disentangled agent for two different multi-task IL environments. In both cases, we out-performed the SOTA by 30% in task success rate. We also demonstrated this for navigation on a real robot.

Instance segmentation is an important task for supporting robotic grasping in auto-store scenarios. Accurate segmentation usually relies on the quantity and quality of available annotated training data. However, it requires tremendous cost to obtain these labels. In this work, without requiring any human annotations on real data, our proposed self-ensembling sim-to-real network, namely SESR, is able to generate precise instance masks for a wide variety of supermarket goods. We design our SESR with a teacher model and a student model trained with a self-ensembling strategy. We adopt different levels of consistency to bridge the sim-to-real gap and boost the model generalization ability. Also, we compile an auto-store bin-picking dataset covering various goods. Extensive experiments on both unseen scenarios and unseen objects validate the effectiveness and superiority of our method over others, and the robot arm demonstrations further show that our segmentation results can support real-time auto-store bin picking.

In this paper, we present the Sample Efficient Social Navigation from Observation (SESNO) algorithm that efficiently learns socially-compliant navigation policies from observations of human trajectories. SESNO is an inverse reinforcement learning (IRL)-based algorithm that learns from human trajectory observations without knowledge of their actions. We improve the sample-efficiency over previous IRL-based methods by introducing a shared experience replay buffer that allows reuse of past trajectory experiences to estimate the policy and the reward. We evaluate SESNO using publicly available pedestrian motion data sets and compare its performance to related baseline methods in the literature. We show that SESNO yields performance superior to existing baselines while dramatically improving the sample complexity by using as few as a hundredth of the samples required by existing baselines.

We propose the framework of Series Elastic End Effectors in 6D (SEED), which combines a spatially compliant element with visuotactile sensing to grasp and manipulate tools in the wild. Our framework generalizes the benefits of series elasticity to 6-dof, while providing an abstraction of control using visuotactile sensing. We propose an algorithm for relative pose estimation from visuotactile sensing, and a spatial hybrid force-position controller capable of achieving stable force interaction with the environment. We demonstrate the effectiveness of our framework on tools that require regulation of spatial forces. Video link: https://youtu.be/2-YuIfspDrk.

This paper introduces SCALER, a quadrupedal robot that demonstrates climbing on bouldering walls, over-hangs, ceilings and trotting on the ground. SCALER is one of the first high-degrees of freedom four-limbed robots that can free-climb under the Earth's gravity and one of the most mechanically efficient quadrupeds on the ground. Where other state-of-the-art climbers specialize in climbing, SCALER promises practical free-climbing with payload and ground locomotion, which realizes true versatile mobility. A new climbing gait, SKATE gait, increases the payload by utilizing the SCALER body linkage mechanism. SCALER achieves a maximum normalized locomotion speed of 1.87 /s, or 0.56 m/s on the ground and 1.0 /min, or 0.35 m/min in bouldering wall climbing. Payload capacity reaches 233 % of the SCALER weight on the ground and 35 % on the vertical wall. Our GOAT gripper, a mechanically adaptable underactuated two-finger gripper, successfully grasps convex and non-convex objects and supports SCALER.

We propose a new SLAM system that uses the semantic segmentation of objects and structures in the scene. Semantic information is relevant as it contains high level information which may make SLAM more accurate and robust. Our contribution is twofold: i) A new SLAM system based on ORB-SLAM2 that creates a semantic map made of clusters of points corresponding to objects instances and structures in the scene. ii) A modification of the classical Bundle Adjustment formulation to constrain each cluster using geometrical priors, which improves both camera localization and reconstruction and enables a better understanding of the scene. We evaluate our approach on sequences from several public datasets and show that it improves camera pose estimation with respect to state of the art.

Autonomous robots are often required to acquire high-level prior knowledge by continuously reconstructing the semantics and geometry of the surrounding scene, which is the basis of exploration and planning. Most existing continuous semantic mapping algorithms cannot distinguish potential differences in voxels, resulting in an over-inflated map. Furthermore, fixed-size query ranges introduce high computational complexity. Based on the limitation of over-inflation and inefficiency, this paper proposes a novel incremental continuous semantic occupancy mapping algorithm (S-MKI). The key innovation of this work comes from the two models in the preprocessing stage. On the one hand, Redundant Voxel Filter Model utilizes context entropy to filter out redundant voxels to improve the confidence of the final map, where objects have accurate boundaries with sharp edges. On the other hand, Adaptive Kernel Length Model adaptively adjusts the kernel length with class entropy, which reduces the inherent amount of training data. The final multientropy kernel inference function is formulated to integrate these two models to infer sparse noisy sensor data into dense accurate 3D maps. Experimental results conducted in both indoors and outdoors datasets validate that S-MKI outperforms existing methods.

Aerial Modular Reconfigurable Robots (AMRRs) are scalable systems consisting of rotor modules capable of rearrangement during flight. The potential to dynamically change any shape for a given task poses the question: what arrangements offer the most aerodynamic benefit for the task of flying? Answering this requires understanding how adjacent rotors in various configurations influence each another. Intuitively, aerodynamic models such as momentum theory suggest that close rotor proximity decreases performance due to the upstream rotor flow fields interacting. However, effects such as vortex interaction or viscous flow entrainment (used by the Dyson bladeless fan) may offer benefits not captured by the modelling assumptions of computational analysis or simulation. Thus, this work takes an experimental approach, testing thrust performance of rotors in independent configurations of lines, square lattices, and hexagons with various inter-rotor spacings. It was found that inter-rotor spacing did not significantly change thrust performance, but that hexagonal arrangements outperformed line and grid lattices. Smoke tests indicated that hexagon configurations entrained air in the central cavity resulting in a thrust improvement. An inter-rotor spacing of 1.51 rotor diameters gave the best performance increase, roughly equal to that of an additional rotor. This suggests that by placing rotors in an array of six hollow hexagonal honeycombs, thrust performance could theoretically be increased by up to 27.3 per cent, for no additional mass.

We develop a method for synthesizing control policies for stochastic, linear, time-varying systems that must perform tasks specified in signal temporal logic. We build upon an efficient, sampling-based framework that computes the probability of the system satisfying its specification. By exploiting the properties of linear systems and robustness score in temporal logic specifications, we obtain sample-efficient gradients of the satisfaction probability with respect to con-troller parameters. Therefore, by applying gradient descent we obtain locally optimized controllers that maximize the chances of satisfying the specification. We demonstrate our approach through examples of a mobile robot and a mobile manipulator in simulation.

In this paper, we describe a robust multi-drone planning framework for high-speed trajectories in large scenes. It uses a free-space-oriented map to free the optimization from cumbersome environment data. A capsule-like safety constraint is designed to avoid reciprocal collisions when vehicles deviate from their nominal flight progress under disturbance. We further show the minimum-singularity differential flatness of our drone dynamics with nonlinear drag effects involved. Leveraging the flatness map, trajectory optimization is efficiently conducted on the flat outputs while still subject to physical limits considering drag forces at high speeds. The robustness and effectiveness of our framework are both validated in large-scale simulations. It can compute collision-free trajectories satisfying high-fidelity vehicle constraints for hundreds of drones within 10 minutes.

Autonomous surgical robotics is a growing area of research, with advances being made in the areas of vision and control. Central to this research is the need for simulations to facilitate data collection and simulate learning environments for Reinforcement Learning (RL) agents. Recent simulators have facilitated RL policy generation, but lack a robust sim2real pipeline and a proven vision-based policy that can use any type of camera including the da Vinci Surgical System (dVSS) Endoscope. To solve this, we build a ROS-based sim2real pipeline that incorporates a Unity3D da Vinci Research Kit (dVRK) simulation, modular kinematics, and shared interfaces. We examine the vision-based task of cube pushing, and train RL policies to execute in real life through Domain Randomization. Our experiments evaluate model success in simulation and two camera systems: OAK-1 and the dVSS Endoscope. Our results indicate that Domain Randomization is effective at bridging the sim2real gap, and even extends to the difficult endoscope scenario. We achieve 100% transfer success rate on both OAK-1 and the dVSS Endoscope, with gains of over 60% compared to a base model with no Domain Randomization. We examine the various randomization parameters, including lighting, camera, and physics variables, and determine that all parameters play a significant role in bridging the sim2real gap. Testing across extreme lighting and camera configurations not seen in simulation, our models continue to perform well, with 85% accuracy on the OAK-1 camera. Our future work will extend to other tasks and more complex policies to take advantage of stereo-camera imaging. Further project information is available at https://medcvr.utm.utoronto.ca/iros2022-sim2real.html

For most LiDAR-inertial odometry, accurate initial states, including temporal offset and extrinsic transfor-mation between LiDAR and 6-axis IMUs, play a significant role and are often considered as prerequisites. However, such information may not be always available in customized LiDAR-inertial systems. In this paper, we propose LI-Init: a full and real-time LiDAR-inertial system initialization process that calibrates the temporal offset and extrinsic parameter between LiDARs and IMUs, and also the gravity vector and IMU bias by aligning the state estimated from LiDAR measurements with that measured by IMU. We implement the proposed method as an initialization module, which can automatically detects the degree of excitation of the collected data and calibrate, on-the-fly, the temporal offset, extrinsic, gravity vector, and IMU bias, which are then used as high-quality initial state values for real-time LiDAR-inertial odometry systems. Experiments conducted with different types of LiDARs and LiDAR-inertial combinations show the robustness, adaptability and efficiency of our initialization method. The implementation of our LiDAR-inertial initialization procedure LI-Init and test data are open-sourced on Github11https://www.github.com/hku-mars/LiDAR IMU Init and also integrated into a state-of-the-art LiDAR-inertial odometry system FAST-LIO2.

Robust localization in a given map is a crucial component of most autonomous robots. In this paper, we address the problem of localizing in an indoor environment that changes and where prominent structures have no correspondence in the map built at a different point in time. To overcome the discrepancy between the map and the observed environment caused by such changes, we exploit human-readable localization cues to assist localization. These cues are readily available in most facilities and can be detected using RGB camera images by utilizing text spotting. We integrate these cues into a Monte Carlo localization framework using a particle filter that operates on 2D LiDAR scans and camera data. By this, we provide a robust localization solution for environments with structural changes and dynamics by humans walking. We evaluate our localization framework on multiple challenging indoor scenarios in an office environment. The experiments suggest that our approach is robust to structural changes and can run on an onboard computer. We release an open source implementation of our approach11https://github.com/PRBonn/tmcl, which uses off-the-shelf text spotting, written in C++ with a ROS wrapper.

When robots walk on uneven terrain, trajectory planning should take into account both the whole-body dy-namics and the ground geometry simultaneously. In uneven terrain environments, there are only a limited number of places where the robot is able to make stable contact with the ground without its feet wobbling or slipping because of the intricate round geometry. In such environments, the optional landing position and time to maintain the robot's balance and stable foot contact are not obvious and computationally expensive. In this study, we propose a robust walking system that integrates environment recognition using steppable regions and walking control for a humanoid robot to walk on uneven terrain. In this paper, a steppable region is defined as a two-dimensional convex hull that represents a region where a robot is capable of landing. We propose a method to compute the steppable region quickly by 2.SD projection of the environment points and spatial filtering. In this system, the walking controller integrates the steppable region with the Capture Region to modify the landing position from a two-dimensional geometric calculation. In addition, to cope with the environment recognition error, we have introduced a trajectory generation that allows the feet to penetrate the ground and hybrid control of position and torque. We verified the effectiveness of the proposed system through experiments in which a life-size humanoid robot walked on uneven terrain and recovered when pushed.

Comprehending human motion is a fundamental challenge for developing Human-Robot Collaborative applications. Computer vision researchers have addressed this field by only focusing on reducing error in predictions, but not taking into account the requirements to facilitate its implementation in robots. In this paper, we propose a new model based on Transformer that simultaneously deals with the real time 3D human motion forecasting in the short and long term. Our 2-Channel Transformer (2CH-TR) is able to efficiently exploit the spatio-temporal information of a shortly observed sequence (400ms) and generates a competitive accuracy against the current state-of-the-art. 2CH-TR stands out for the efficient performance of the Transformer, being lighter and faster than its competitors. In addition, our model is tested in conditions where the human motion is severely occluded, demonstrating its robustness in reconstructing and predicting 3D human motion in a highly noisy environment. Our experiment results show that the proposed 2CH-TR outperforms the ST-Transformer, which is another state-of-the-art model based on the Transformer, in terms of reconstruction and prediction under the same conditions of input prefix. Our model reduces in 8.89% the mean squared error of ST-Transformer in short-term prediction, and 2.57% in long-term prediction in Human3.6M dataset with 400ms input prefix.

Deep reinforcement learning has emerged as a popular and powerful way to develop locomotion controllers for quadruped robots. Common approaches have largely focused on learning actions directly in joint space, or learning to modify and offset foot positions produced by trajectory generators. Both approaches typically require careful reward shaping and training for millions of time steps, and with trajectory generators introduce human bias into the resulting control policies. In this paper, we present a learning framework that leads to the natural emergence of fast and robust bounding policies for quadruped robots. The agent both selects and controls actions directly in task space to track desired velocity commands subject to environmental noise including model uncertainty and rough terrain. We observe that this framework improves sample efficiency, necessitates little reward shaping, leads to the emergence of natural gaits such as galloping and bounding, and eases the sim-to-real transfer at running speeds. Policies can be learned in only a few million time steps, even for challenging tasks of running over rough terrain with loads of over 100% of the nominal quadruped mass. Training occurs in PyBullet, and we perform a sim-to-sim transfer to Gazebo and sim-to-real transfer to the Unitree A1 hardware. For sim-to-sim, our results show the quadruped is able to run at over 4 m/s without a load, and 3.5 m/s with a 10 kg load, which is over 83% of the nominal quadruped mass. For sim-to-real, the Unitree A1 is able to bound at 2 m/s with a 5 kg load, representing 42% of the nominal quadruped mass.

Signal temporal logic (STL) provides a powerful, flexible framework for specifying complex autonomy tasks; however, existing methods for planning based on STL specifications have difficulty scaling to long-horizon tasks and are not robust to external disturbances. In this paper, we present an algorithm for finding robust plans that satisfy STL specifications. Our method alternates between local optimization and local falsification, using automatically differentiable temporal logic to iteratively optimize its plan in response to counterexamples found during the falsification process. We benchmark our counterexample-guided planning method against state-of-the-art planning methods on two long-horizon satellite rendezvous missions, showing that our method finds high-quality plans that satisfy STL specifications despite adversarial disturbances. We find that our method consistently finds plans that are robust to adversarial disturbances and requires less than half the time of competing methods. We provide an implementation of our planner at https://github.com/MIT-REALM/architect.

In this article, we propose a deep learning frame-work that provides a unified approach to the problem of leg contact detection in humanoid robot walking gaits. Our formulation accomplishes to accurately and robustly estimate the contact state probability for each leg (i.e., stable or slip/no contact). The proposed framework employs solely propriocep-tive sensing and although it relies on simulated ground-truth contact data for the classification process, we demonstrate that it generalizes across varying friction surfaces and different legged robotic platforms and, at the same time, is readily transferred from simulation to practice. The framework is quantitatively and qualitatively assessed in simulation via the use of ground-truth contact data and is contrasted against state-of-the-art methods with an ATLAS, a NAO, and a TALOS humanoid robot. Furthermore, its efficacy is demonstrated in base estimation with a real TALOS humanoid. To reinforce further research endeavors, our implementation is offered as an open-source ROS/Python package, coined Legged Contact Detection (LCD).

The ability to reason about changes in the environment is crucial for robots operating over extended periods of time. Agents are expected to capture changes during operation so that actions can be followed to ensure a smooth progression of the working session. However, varying viewing angles and accumulated localization errors make it easy for robots to falsely detect changes in the surrounding world due to low observation overlap and drifted object associations. In this paper, based on the recently proposed category-level Neural Descriptor Fields (NDFs), we develop an object-level online change detection approach that is robust to partially overlapping observations and noisy localization results. Utilizing the shape completion capability and SE(3)-equivariance of NDFs, we represent objects with compact shape codes encoding full object shapes from partial observations. The objects are then organized in a spatial tree structure based on object centers recovered from NDFs for fast queries of object neighborhoods. By associating objects via shape code similarity and comparing local object-neighbor spatial layout, our proposed approach demonstrates robustness to low observation overlap and localization noises. We conduct experiments on both synthetic and real-world sequences and achieve improved change detection results compared to multiple baseline methods. Project web-page: ?http://yilundu.github.io/ndf_change