ICRA 2021 论文列表

Aerial manipulation increases significantly the workspace size of robotic manipulators. However, aerial manipulation suffers from a lack of autonomy due to limited embedded energy. The Aerial Manipulator with Elastic Suspension (AMES) is designed to cope with this issue. It is an omnidirectional aerial vehicle equipped with a gripper and suspended under a robotic carrier by a spring for gravity compensation. In this paper, the AMES is controlled with a nonlinear model predictive controller (NMPC). To eliminate the steady-state errors, an observer based on a model of the AMES augmented with constant disturbances is implemented in conjunction with the NMPC controller. Experiments illustrate the efficiency of the NMPC by comparing it to a computed torque controller.

We present non-convex maximal dissipation principle (NMDP), a time integration scheme for articulated bodies with simultaneous contacts. Our scheme resolves contact forces via the maximal dissipation principle (MDP). Whereas prior MDP solvers assume linearized dynamics and integrate using the forward multistep scheme, we consider the coupled system of nonlinear Newton-Euler dynamics and MDP and integrate using the backward integration scheme. We show that the coupled system of equations can be solved efficiently using a novel projected gradient method with guaranteed convergence. We evaluate our method by predicting several locomotion trajectories for a quadruped robot. The results show that our NMDP scheme has several desirable properties including: (1) generalization to novel contact models; (2) stability under large timestep sizes; (3) consistent trajectory generation under varying timestep sizes.

This paper aims at showing the dynamic performance and reliability of the low-cost, open-access quadruped robot Solo-12, which is developed within the framework of Open Dynamic Robot Initiative. It presents the implementation of a state-of-the-art control pipeline, close to the one that was previously implemented on Mini Cheetah, which implements a model predictive controller based on the centroidal dynamics to compute desired contact forces in order to track a reference velocity. Different contributions are proposed to speed up the computation process, notably at the level of the state estimation and the whole body controller. Experimental results demonstrate that the robot closely follow the reference velocity while being highly reactive and able to recover from perturbations.

Impact mitigation is crucial to the stable locomotion of legged robots, especially in high-speed dynamic locomotion. This paper presents a leg locomotion system, including the nonlinear active compliance control and the active impedance control for the steel wire transmission-based legged robot. The developed control system enables high-speed dynamic locomotion with excellent impact mitigation and leg position tracking performance, where three strategies are applied. a) The feed-forward controller is designed according to the linear motor-leg model with the information of Coulomb friction and viscous friction. b) Steel wire transmission model-based compensation guarantees ideal virtual spring compliance characteristics. c) Nonlinear active compliance control and active impedance control ensure better impact mitigation performance than linear scheme and guarantee position tracking performance. The proposed control system is verified on a real robot named SCIT Dog, and the experiment demonstrates the ideal impact mitigation ability in high-speed dynamic locomotion without any passive spring mechanism.

Aleatoric uncertainty estimation, based on the observed training data, is applied for the detection of conflicts in a demonstration data set. The particular focus of this paper is the resolution of conflicting data resulting from scenarios with equivalent action choices, such as obstacle avoidance, path planning or multiple joint configurations. In terms of the estimated uncertainty, the proposed algorithm aims to decrease this otherwise irreducible value through direct alteration of the accrued data set and to provide data that a policy-learning neural network is able to fit appropriately. The proposed algorithm was validated with real robot scenarios while learning from inconsistent demonstrations, where the resulting policies consistently achieved their prescribed objectives. A video showing our method and experiments can be found at: https://youtu.be/oGYnzlW9Ncw.

We present a learning algorithm for training a single policy that imitates multiple gaits of a walking robot. To achieve this, we use and extend MPC-Net, which is an Imitation Learning approach guided by Model Predictive Control (MPC). The strategy of MPC-Net differs from many other approaches since its objective is to minimize the control Hamiltonian, which derives from the principle of optimality. To represent the policies, we employ a mixture-of-experts network (MEN) and observe that the performance of a policy improves if each expert of a MEN specializes in controlling exactly one mode of a hybrid system, such as a walking robot. We introduce new loss functions for single- and multi-gait policies to achieve this kind of expert selection behavior. Moreover, we benchmark our algorithm against Behavioral Cloning and the original MPC implementation on various rough terrain scenarios. We validate our approach on hardware and show that a single learned policy can replace its teacher to control multiple gaits.

In recent years, mobility on demand has experienced a major revival due to various ride-hailing companies entering the market. Competing in this field requires an efficient operation. Therefore, the applied policy, which cares for vehicle-to-customer assignment and vehicle repositioning, has to achieve good customer service and minimize cost while trying to keep the impact on the environment as low as possible. A promising approach is to coordinate the control of the entire fleet, which is foreseen to become even easier with the possibility of autonomous vehicles in mind. Anticipating future demand requires a good understanding of the spatiotemporal distributions of request origins and destinations, and the resulting imbalance between vehicle demand and availability. This results from a multitude of topological, demographic, and social effects, which are almost impossible to sufficiently capture in a handcrafted model of reasonable complexity. This can be circumvented by leveraging machine learning approaches. In this paper, an image-like representation of the city and its fleet's state is introduced. It is comprehensive and intuitive to use as input to convolutional neural networks, which in the past have already been proven to capture spatial relationships very well. This allows operating on realistic, full-sized traffic networks without greatly increasing the number of parameters the neural network has to learn and, hence, keeps the training effort low. Additionally, this state is combined with a similarly constructed repositioning action, reflecting a 2D distribution of a well-performing operational policy. This approach allows replacement of complex, handcrafted mathematical models by a single, compact, auto-encoder-like neural network.

The ability to detect and estimate external contacts is essential for robot arms to operate in unstructured environments occupied by humans. However, most robot arms are not equipped with adequate sensors to detect contacts on their entire body. What many robot arms do have is torque sensors for individual joints. Through a quantitative analysis, we argue that it is fairly likely for two distinct contacts on the robot’s surface to generate almost identical joint torque measurements. When this happens, the best contact estimate achievable is the set of possible contact positions, all of which would reproduce the measured joint torque. Searching for elements of this set is equivalent to solving to global optimality a nonlinear program.By combining rejection sampling with gradient descent, we propose a contact estimation method which in practice finds all local optima of the nonlinear program at real-time rates. In addition, we propose an active contact exploration method which falsifies spurious contact estimates in the set of local optima by making small motions around the robot’s current configuration. The proposed methods highlight the caveats of contact estimation from only joint torque, which, coupled with known limitations of such estimators, suggest that a more capable sensor is probably needed for robust whole-body contact estimation.

Interactive driving scenarios, such as lane changes, merges and unprotected turns, are some of the most challenging situations for autonomous driving. Planning in interactive scenarios requires accurately modeling the reactions of other agents to different future actions of the ego agent. We develop end-to-end models for conditional behavior prediction (CBP) that take as an input a query future trajectory for an ego-agent, and predict distributions over future trajectories for other agents conditioned on the query. Leveraging such a model, we develop a general-purpose agent interactivity score derived from probabilistic first principles. The interactivity score allows us to find interesting interactive scenarios for training and evaluating behavior prediction models. We further demonstrate that the proposed score is effective for agent prioritization under computational budget constraints.

The localization system, which outputs vehicle position, velocity, and attitude, is one of the fundamental components in the autonomous driving vehicle. The global pose is not only used for the planning and control system, but also an important reference for the cloud source-based HD Map building and updating. The accuracy, availability, and reliability are key requirements for the localization system to ensure that the whole system runs smoothly and efficiently.IMU/Vehicle extrinsic calibration is one of the primary jobs that should be addressed. Due to the observability issue, the IMU/vehicle relative roll cannot be calibrated by the traditional maneuver-based calibration method. In this paper, we solve this issue with the proposed Multiple Orientation-based Vehicle/IMU Extrinsic Calibration (MOVIE-Cali) method, which is evaluated by Monte Carlo simulations and experiments.When the vehicle is cornering or making a U-turn, the sideslip of the tires will have negative influence on the localization system which uses Non-Holonomic Constraints (NHC)/Wheel speed sensor measurement in the model. We derive a sideslip angle model and propose an online slip parameter calibration and compensation method to improve the localization accuracy. The performance of proposed method has been evaluated by the vehicle tests.

In this work, we propose a novel method for performing inertial aided navigation, by using deep neural net-works (DNNs). To date, most DNN inertial navigation methods focus on the task of inertial odometry, by taking gyroscope and accelerometer readings as input and regressing for integrated IMU poses (i.e., position and orientation). While this design has been successfully applied on a number of applications, it is not of theoretical performance guarantee unless patterned motion is involved. This inevitably leads to significantly reduced accuracy and robustness in certain use cases. To solve this problem, we design a framework to compute observable IMU integration terms using DNNs, followed by the numerical pose integration and sensor fusion to achieve the performance gain. Specifically, we perform detailed analysis on the motion terms in IMU kinematic equations, propose a dedicated network design, loss functions, and training strategies for the IMU data processing, and conduct extensive experiments. The results show that our method is generally applicable and outperforms both traditional and DNN methods by wide margins.

The IKEA Furniture Assembly Environment is one of the first benchmarks for testing and accelerating the automation of long-horizon and hierarchical manipulation tasks. The environment is designed to advance reinforcement learning and imitation learning from simple toy tasks to complex tasks requiring both long-term planning and sophisticated low-level control. Our environment features 60 furniture models, 6 robots, photorealistic rendering, and domain randomization. We evaluate reinforcement learning and imitation learning methods on the proposed environment. Our experiments show furniture assembly is a challenging task due to its long horizon and sophisticated manipulation requirements, which provides ample opportunities for future research. The environment is publicly available at https://clvrai.com/furniture.

Assistive robots have the potential to help people perform everyday tasks. However, these robots first need to learn what it is their user wants them to do. Teaching assistive robots is hard for inexperienced users, elderly users, and users living with physical disabilities, since often these individuals are unable to show the robot their desired behavior. We know that inclusive learners should give human teachers credit for what they cannot demonstrate. But today’s robots do the opposite: they assume every user is capable of providing any demonstration. As a result, these robots learn to mimic the demonstrated behavior, even when that behavior is not what the human really meant! Here we propose a different approach to reward learning: robots that reason about the user’s demonstrations in the context of similar or simpler alternatives. Unlike prior works — which err towards overestimating the human’s capabilities — here we err towards underestimating what the human can input (i.e., their choice set). Our theoretical analysis proves that underestimating the human’s choice set is riskaverse, with better worst-case performance than overestimating. We formalize three properties to generate similar and simpler alternatives. Across simulations and a user study, our resulting algorithm better extrapolates the human’s objective. See the user study here: https://youtu.be/RgbH2YULVRo.

In this work, we present an adaptive behavior path planning method for autonomous exploration and visual search of unknown environments. As volumetric exploration and visual coverage of unknown environments, with possibly different sensors, are non-identical objectives, a principled combination of the two is proposed. In particular, the method involves three distinct planning policies, namely exploration, and sparse or dense visual coverage. A hypergame formulation is proposed which allows the robot to select for the next-best planning behavior in response to the currently encountered environment challenges in terms of geometry and visual conditions, alongside a self-assessment of its performance. The proposed planner is evaluated in a collection of experimental and simulation studies in diverse environments, while comparative results against a state-of-the-art exploration method are also presented.

We address the problem of image registration to a compressed 3D map. While this is most often performed by comparing LiDAR scans to the point cloud based map, it depends on an expensive LiDAR sensor at run time and the large point cloud based map creates overhead in data storage and transmission. Recently, efforts have been underway to replace the expensive LiDAR sensor with cheaper cameras and perform 2D-3D localization. In contrast to the previous work that learns relative pose by comparing projected depth and camera images, we propose HyperMap, a paradigm shift from online depth map feature extraction to offline 3D map feature computation for the 2D-3D camera registration task through end-to-end training. In the proposed pipeline, we first perform offline 3D sparse convolution to extract and compress the voxelwise hypercolumn features for the whole map. Then at run-time, we project and decode the compressed map features to the rough initial camera pose to form a virtual feature image. A Convolutional Neural Network (CNN) is then used to predict the relative pose between the camera image and the virtual feature image. In addition, we propose an efficient occlusion handling layer, specifically designed for large point clouds, to remove occluded points in projection. Our experiments on synthetic and real datasets show that, by moving the feature computation load offline and compressing, we reduced map size by 87−94% while maintaining comparable or better accuracy.

This study proposes a hybrid vision/force control scheme for interaction with the inner surface of the bottle-like object. Based on the geometry of the object, a new generalized constraint called the bottleneck (BN) constraint is proposed, which ensures the tool passes through a fixed 3-D region and avoid collisions with the boundary of the region. To realize the hybrid vision/force control under the BN constraint, a novel dynamic controller is designed inspired by the hierarchical operational space, which can complete the different tasks defined in the decoupled subspace. To enhance the robustness of the algorithm, we develop a data-driven method and an adaptive method to estimate the Jacobian matrix online in force space and image space, respectively. The asymptotic stability of the closed-loop system is rigorously proved by the Lyapunov theory. Experiments are conducted to validate the performance of the proposed method.

This paper proposes a hybrid planning framework that generates complex dynamic motion plans for jumping legged robots to traverse challenging terrains. By employing a motion primitive, the original problem is decoupled as path planning followed by a trajectory optimization (TO) module that handles dynamics. A variant of a kinodynamic Rapidly-exploring Random Trees (RRT) planner finds a path as a parabola sequence between stance phases. To make this fast, a reachability informed control sampling scheme leverages a precomputed velocity reachability map. The path is post-processed to eliminate redundant jumps and passed to the TO module to find a dynamically feasible trajectory. Simulation results are presented where the proposed hybrid planner solves challenging terrains by executing multiple consecutive jumps, producing novel strategies to leap over large gaps by leveraging dynamics. In a physical experiment, the hybrid planner is tested on a real robot successfully traversing a challenging terrain.

Powered exoskeletons for people with paraplegia have been widely developed. To generate the basic but essential motions for daily human life, precise control algorithms to follow the joint reference trajectories are necessary. The dynamic characteristics of the exoskeletal joints, however, varies signifi-cantly during walking because the load side is exchanged from legs in the air to the wearer’s body. To ensure robustness and tracking performance for any case of gait even in the presence of exogenous disturbances such as human’s active movements and repeated ground contacts, customized robust control algorithms need to be developed. In this paper, therefore, hybrid model control of the powered exoskeleton, WalklON Suit, utilized with the disturbance observer is introduced. A hybrid nominal model, whose model parameters are interchanged between the gait phases, i.e., swing and stance, is developed by the parameter adatpation algorithm. By the proposed method, the disturbance observer can fully reject the exogenous disturbance during walking and achieve high-performance gait assistance to the people with complete paraplegia. In this paper, the experimental verification of the designed model and the controller with the WalkON Suit, are also introduced.

Vision-based localization and mapping solution is promising to be adopted in the automated valet parking task. In this paper, a semantic SLAM framework that leverages the hybrid edge information on bird’s-eye view images is presented. To extract useful edges from the synthesized bird’s-eye view image and the free-space contours for the SLAM task, different segmentation methods are designed to remove the noisy glare edges and distorted object edges caused by the inverse perspective mapping in view synthesis. Since only the free-space segmentation model needs training, our methods can dramatically reduce the labeling burden compared with previous road marking based methods. Those incorrect and incomplete edges are further cleaned and recovered by a temporal fusion of consecutive edges in a local map, respectively. Both a semantic edge point cloud map and an occupancy grid map can be built simultaneously in real time. Experiments in a parking garage demonstrate that the proposed framework can achieve higher accuracy and perform more robustly than previous point feature based methods.