ICRA 2020 论文列表

We present an algorithm for the concurrent assignment and planning of collision-free trajectories (DC-CAPT) for robots whose kinematics can be modeled as Dubins cars, i.e., robots constrained in terms of their initial orientation and their minimum turning radius. Coupling the assignment and trajectory planning subproblems allows for a computationally tractable solution. This solution is guaranteed to be collision- free through the use of a single constraint: the start and goal locations have separation distance greater than some threshold. We derive this separation distance by extending a prior work that assumed holonomic robots. We demonstrate the validity of our approach, and show its efficacy through simulations and experiments where groups of robots executing Dubins curves travel to their assigned goal locations without collisions.

Programming languages, libraries, and development tools have transformed the application development processes for mobile computing and machine learning. This paper introduces CyPhyHouse—a toolchain that aims to provide similar programming, debugging, and deployment benefits for distributed mobile robotic applications. Users can develop hardware-agnostic, distributed applications using the high-level, event driven Koord programming language, without requiring expertise in controller design or distributed network protocols. The modular, platform-independent middleware of CyPhyHouse implements these functionalities using standard algorithms for path planning (RRT), control (MPC), mutual exclusion, etc. A high-fidelity, scalable, multi-threaded simulator for Koord applications is developed to simulate the same application code for dozens of heterogeneous agents. The same compiled code can also be deployed on heterogeneous mobile platforms. The effectiveness of CyPhyHouse in improving the design cycles is explicitly illustrated in a robotic testbed through development, simulation, and deployment of a distributed task allocation application on in-house ground and aerial vehicles.

The only way to perceive a small object held between our fingers is to trust our sense of touch. Touch provides cues about the state of the contact even if its view is occluded by the finger. The interaction between the soft fingers and the surface reveals crucial information, such as the local shape of the object, that plays a central role in fine manipulation. In this work, we present a new spherical sensor that endows robots with a fine distributed sense of touch. This sensor is an evolution of our distributed tactile sensor that measures the dense 3-dimensional displacement field of an elastic membrane, using the subtractive color-mixing principle. We leverage a planar manufacturing process that enables the design and manufacturing of the functional features on a flat surface. The flat functional panels are then folded to create a spherical shape able to sense a wide variety of objects.The resulting 40mm-diameter spherical sensor has 77 measurement points, each of which gives an estimation of the local 3d displacement, normal and tangential to the surface. Each marker is built around 2 sets of colored patches placed at different depths. The relative motion and resulting hue of each marker, easily captured by an embedded RGB camera, provides a measurement of their 3d motion. To benchmark the sensor, we compared the measurements obtained while pressing the sensor on a curved surface with Hertz contact theory, a hallmark of contact mechanics. While the mechanics did strictly follow Hertz contact theory, using the shear and normal sensing, ChromaTouch can estimate the curvature of an object after a millimeter-size indentation of the sensor.

Imitation learning enables robots to learn a task by simply watching the demonstration of the task. Current imitation learning methods usually require the learner and demonstrator to occur in the same context. This limits their scalability to practical applications. In this paper, we propose a more general imitation learning method which allows the learner and the demonstrator to come from different contexts, such as different viewpoints, backgrounds, and object positions and appearances. Specifically, we design a robotic system consisting of three models: context translation model, depth prediction model and multi-modal inverse dynamics model. First, the context translation model translates the demonstration to the context of learner from a different context. Then combining the color observation and depth observation as inputs, the inverse model maps the multi-modal observations into actions to reproduce the demonstration, where the depth observation is provided by a depth prediction model. By performing the block stacking tasks both in simulation and real world, we prove the cross-context learning advantage of the proposed robotic system over other systems.

We introduce Crocoddyl (Contact RObot COntrol by Differential DYnamic Library), an open-source framework tailored for efficient multi-contact optimal control. Crocoddyl efficiently computes the state trajectory and the control policy for a given predefined sequence of contacts. Its efficiency is due to the use of sparse analytical derivatives, exploitation of the problem structure, and data sharing. It employs differential geometry to properly describe the state of any geometrical system, e.g. floating-base systems. Additionally, we propose a novel optimal control algorithm called Feasibility-driven Differential Dynamic Programming (FDDP). Our method does not add extra decision variables which often increases the computation time per iteration due to factorization. FDDP shows a greater globalization strategy compared to classical Differential Dynamic Programming (DDP) algorithms. Concretely, we propose two modifications to the classical DDP algorithm. First, the backward pass accepts infeasible state-control trajectories. Second, the rollout keeps the gaps open during the early "exploratory" iterations (as expected in multipleshooting methods with only equality constraints). We showcase the performance of our framework using different tasks. With our method, we can compute highly-dynamic maneuvers (e.g. jumping, front-flip) within few milliseconds.

Correspondence identification is an essential problem for collaborative multi-robot perception, with the objective of deciding the correspondence of objects that are observed in the field of view of each robot. In this paper, we introduce a novel maximin hypergraph matching approach that formulates correspondence identification as a hypergraph matching problem. The proposed approach incorporates both spatial relationships and appearance features of objects to improve representation capabilities. It also integrates the maximin theorem to optimize the worst-case scenario in order to address distractions caused by non-covisible objects. In addition, we design an optimization algorithm to address the formulated non-convex non-continuous optimization problem. We evaluate our approach and compare it with seven previous techniques in two application scenarios, including multi-robot coordination on real robots and connected autonomous driving in simulations. Experimental results have validated the effectiveness of our approach in identifying object correspondence from partially overlapped views in collaborative perception, and have shown that the proposed maximin hypergraph matching approach outperforms previous techniques and obtains state-of-the-art performance.

Despite substantial evidence for the crucial role played by an active backbone or spine in animal locomotion, its adoption in legged robots remains limited because the added mechanical complexity and resulting dynamical challenges pose daunting obstacles to characterizing even a partial range of potential performance benefits. This paper takes a next step toward such a characterization by exploring the quasistatic terrestrial self-righting mechanics of a model system with coronal plane spine twisting (CPST). Reduction from a full 3D kinematic model of CPST to a two parameter, two degree of freedom coronal plane representation of body shape affordance predicts a substantial benefit to ground righting by lowering the barrier between stable potential energy basins. The reduced model predicts the most advantageous twist angle for several cross-sectional geometries, reducing the required righting torque by up to an order of magnitude depending on constituent shapes. Experiments with a three actuated degree of freedom physical mechanism corroborate the kinematic model predictions using two different quasistatic reorientation maneuvers for both elliptical and rectangular shaped bodies with a range of eccentricities or aspect ratios. More speculative experiments make intuitive use of the kinematic model in a highly dynamic maneuver to suggest still greater benefits of CPST achievable by coordinating kinetic as well as potential energy, for example as in a future multi-appendage system interacting with a contact-rich 3D environment.

In this work, we derive a coordinate-free formulation of the coupled dynamics of a class of 6DOF aerial manipulators consisting of an underactuated quadrotor equipped with a 2DOF articulated manipulator, and demonstrate that the system is differentially flat with respect to the end effector pose. In particular, we require the center of mass of the entire system to be fixed in the end effector frame, suggesting a reasonable mechanical design criterion. We make use of an inertial decoupling transformation to demonstrate differential flatness, allowing us to plan dynamically feasible trajectories for the system in the space of the 6DOF pose of the end effector, which is ideal for achieving precise manipulator tasks. Simulation results validate the flatness-based planning methodology for our dynamic model, and its usefulness is demonstrated in a simulated aerial videography task.

Fully autonomous vehicles are expected to share the road with less advanced vehicles for a significant period of time. Furthermore, an increasing number of vehicles on the road are equipped with a variety of low-fidelity sensors which provide some perception and localization data, but not at a high enough quality for full autonomy. In this paper, we develop a perception and localization system that allows a vehicle with low-fidelity sensors to incorporate high-fidelity observations from a vehicle in front of it, allowing both vehicles to operate with full autonomy. The resulting system generates perception and localization information that is both low-noise in regions covered by high-fidelity sensors and avoids false negatives in areas only observed by low-fidelity sensors, while dealing with latency and dropout of the communication link between the two vehicles. At its core, the system uses a set of Extended Kalman filters which incorporate observations from both vehicles' sensors and extrapolate them using information about the road geometry. The perception and localization algorithms are evaluated both in simulation and on real vehicles as part of a full cooperative driving system.

The challenges of multi-robot navigation in dynamic environments lie in uncertainties in obstacle complexities, partially observation of robots, and policy implementation from simulations to the real world. This paper presents a cooperative approach to address the multi-robot navigation problem (MRNP) under dynamic environments using a deep reinforcement learning (DRL) framework, which can help multiple robots jointly achieve optimal paths despite a certain degree of obstacle complexities. The novelty of this work includes threefold: (1) developing a cooperative architecture that robots can exchange information with each other to select the optimal target locations; (2) developing a DRL based framework which can learn a navigation policy to generate the optimal paths for multiple robots; (3) developing a training mechanism based on dynamics randomization which can make the policy generalized and achieve the maximum performance in the real world. The method is tested with Gazebo simulations and 4 differential drive robots. Both simulation and experiment results validate the superior performance of the proposed method in terms of success rate and travel time when compared with the other state-of-art technologies.

Cooperative autonomy and data sharing can largely improve the mission performance of robotic networks in underwater surveillance applications. In this paper, we describe the cooperative autonomy used to control the Autonomous Underwater Vehicles (AUVs) acting as sonar receiver nodes in the CMRE Anti-Submarine Warfare (ASW) network. The paper focuses on a track management module that was integrated in the robot autonomy software for enabling the share of information. Track to track (T2T) associations are used for improving track classification and for creating a common tactical picture, necessary for AUV cooperative strategies. We also present a new cooperative data-driven AUV behaviour that exploits the spatial diversity of multiple robots for improving target tracking and for facilitating T2T associations. We report results with real data collected at sea that validate the approach. The reported results are one of the first examples that show the potential of cooperative autonomy and data fusion in realistic underwater surveillance scenarios characterised by limited communications.

Implementing an intuitive control law for an upper-limb exoskeleton dedicated to force augmentation is a challenging issue in the field of human-robot collaboration. The aim of this study is to design an innovative approach to assist carrying an unknown load without using force sensors or specific handle. The method is based on user's intentions estimated through a wireless EMG armband allowing movement direction and intensity estimation along 1 Degree of Freedom. This control law aimed to behave like a gravity compensation except that the mass of the load does not need to be known. The proposed approach was tested on 10 participants during a lifting task with a single Degree of Freedom upper-limb exoskeleton. Participants performed it in three different conditions : without assistance, with an exact gravity compensation and with the proposed method based on EMG armband. The evaluation of the efficiency of the assistance was based on EMG signals captured on seven muscles (objective indicator) and a questionnaire (subjective indicator). Results showed a statically significant reduction of mean activity of the biceps, erector spinae and deltoid by 20%±14%, 18%±12% and 25% ± 16% respectively while comparing the proposed method with no assistance. In addition, similar muscle activities were found both in the proposed method and the traditional gravity compensation. Subjective evaluation showed better precision, efficiency and responsiveness of the proposed method compared to the traditional one.

Hybrid wheel-legged robots have begun to demonstrate the ability to adapt to complex terrain traditionally inaccessible to purely wheeled morphologies. Further research is needed into how their dynamics can be optimally controlled for developing highly adaptive behaviours on challenging terrain. Using optimal center of mass (COM) kinematic trajectories, this work examines the nonlinear dynamics control problem for fast height adaptation on the hybrid humanoid platform known as Aerobot. We explore the dynamics control problem through experimentation with an offline trajectory optimisation (TO) method and a task-space inverse dynamics (TSID) controller for varying the robot's height. Our TO approach uses sequential quadratic programming (SQP) to solve optimal 7th order spline coefficients for the robot's kinematics. The nonlinear Zero Moment Point (ZMP) is used to model a stability criterion that is constrained in the TO problem to ensure dynamic stability. Our TSID controller follows motion plans based on using task jacobians and a simplified passive dynamics model of the Aerobot platform. Results exhibit fast height adaptation on the Aerobot platform with significantly differing results between the control methods that prompts new research into how it may be controlled online.

Assistive robotic arms enable users with physical disabilities to perform everyday tasks without relying on a caregiver. Unfortunately, the very dexterity that makes these arms useful also makes them challenging to teleoperate: the robot has more degrees-of-freedom than the human can directly coordinate with a handheld joystick. Our insight is that we can make assistive robots easier for humans to control by leveraging latent actions. Latent actions provide a lowdimensional embedding of high-dimensional robot behavior: for example, one latent dimension might guide the assistive arm along a pouring motion. In this paper, we design a teleoperation algorithm for assistive robots that learns latent actions from task demonstrations. We formulate the controllability, consistency, and scaling properties that user-friendly latent actions should have, and evaluate how different lowdimensional embeddings capture these properties. Finally, we conduct two user studies on a robotic arm to compare our latent action approach to both state-of-the-art shared autonomy baselines and a teleoperation strategy currently used by assistive arms. Participants completed assistive eating and cooking tasks more efficiently when leveraging our latent actions, and also subjectively reported that latent actions made the task easier to perform. The video accompanying this paper can be found at: https://youtu.be/wjnhrzugBj4.

The interplay between network topology and the interaction modalities of a multi-robot team fundamentally impact the types of formations that can be achieved. To explore the trade-offs between network structure and the sensing and communication capabilities of individual robots, this paper applies controller synthesis to formation control of infinitesimally shape-similar frameworks, for which maintaining the relative angles between robots ensures invariance of the framework to translation, rotation, and uniform scaling. Beginning with the development of a controller for the sole purpose of maintaining the formation, the controller-synthesis approach is introduced as a mechanism for incorporating user- designated objectives while ensuring that the formation is maintained. Both centralized and decentralized formulations of the synthesized controller are presented, the resulting sensing and communication requirements are discussed, and the method is demonstrated on a team of differential-drive robots.

The paper deals with non-standard motion tasks specified for a two-wheeled nonholonomic robot. It is assumed that wheels cannot fully rotate which reduces a set of feasible movements significantly. In spite of these constraints, it is expected that position of the robot can be changed without violating nonholonomic constraints. Such a possibility comes from the small time local controllability (STLC) of the kinematics described on four-dimensional configuration manifold. In order to solve these specific tasks a feedback taking advantage of the transverse function approach is designed. Consequently, the system can be virtually released from non-holonomic constraints. The transverse function also defines a virtual geometry constraint which makes it possible to limit wheels rotation.Properties of the designed controller are illustrated by results of numerical simulations in various motion task scenarios.

We present a reinforcement learning (RL) frame-work to synthesize a control policy from a given linear temporal logic (LTL) specification in an unknown stochastic environment that can be modeled as a Markov Decision Process (MDP). Specifically, we learn a policy that maximizes the probability of satisfying the LTL formula without learning the transition probabilities. We introduce a novel rewarding and discounting mechanism based on the LTL formula such that (i) an optimal policy maximizing the total discounted reward effectively maximizes the probabilities of satisfying LTL objectives, and (ii) a model-free RL algorithm using these rewards and discount factors is guaranteed to converge to such a policy. Finally, we illustrate the applicability of our RL-based synthesis approach on two motion planning case studies.

This paper presents a method to enable a robot using stochastic Model Predictive Control (MPC) to achieve high performance on a repetitive path-following task. In particular, we consider the case where the accuracy of the model for robot dynamics varies significantly over the path-motivated by the fact that the models used in MPC must be computationally efficient, which limits their expressive power. Our approach is based on correcting the cost predicted using a simple learned dynamics model over the MPC horizon. This discourages the controller from taking actions that lead to higher cost than would have been predicted using the dynamics model. In addition, stochastic MPC provides a quantitative measure of safety by limiting the probability of violating state and input constraints over the prediction horizon. Our approach is unique in that it combines both online model learning and cost learning over the prediction horizon and is geared towards operating a robot in changing conditions. We demonstrate our algorithm in simulation and experiment on a ground robot that uses a stereo camera for localization.

An essential measure of autonomy in assistive service robots is adaptivity to the various contexts of human-oriented tasks, which are subject to subtle variations in task parameters that determine optimal behaviour. In this work, we propose an apprenticeship learning approach to achieving context-aware action generalization on the task of robot-to-human object hand-over. The procedure combines learning from demonstration and reinforcement learning: a robot first imitates a demonstrator's execution of the task and then learns contextualized variants of the demonstrated action through experience. We use dynamic movement primitives as compact motion representations, and a model-based C-REPS algorithm for learning policies that can specify hand-over position, conditioned on context variables. Policies are learned using simulated task executions, before transferring them to the robot and evaluating emergent behaviours. We additionally conduct a user study involving participants assuming different postures and receiving an object from a robot, which executes hand-overs by either imitating a demonstrated motion, or adapting its motion to hand-over positions suggested by the learned policy. The results confirm the hypothesized improvements in the robot's perceived behaviour when it is context-aware and adaptive, and provide useful insights that can inform future developments.