ICRA 2020 论文列表

We present a contact-implicit trajectory optimization framework that can plan contact-interaction trajectories for different robot architectures and tasks using a trivial initial guess and without requiring any parameter tuning. This is achieved by using a relaxed contact model along with an automatic penalty adjustment loop for suppressing the relaxation. Moreover, the structure of the problem enables us to exploit the contact information implied by the use of relaxation in the previous iteration, such that the solution is explicitly improved with little computational overhead. We test the proposed approach in simulation experiments for non-prehensile manipulation using a 7-DOF arm and a mobile robot and for planar locomotion using a humanoid-like robot in zero gravity. The results demonstrate that our method provides an out-of-the-box solution with good performance for a wide range of applications.

Soft robots demonstrate great potential compared with traditional rigid robots owing to their inherently soft body structures. Although researchers have made tremendous progress in recent years, existing soft robots are in general plagued by a main issue: slow speeds and small forces. In this work, we aim to address this issue by actively designing the energy landscape of the soft body: the total strain energy with respect to the robot's deformation. With such a strategy, a soft robot's dynamics can be tuned to have fast and strong motion. We introduce the general design principle using a soft module with two stable states that can rapidly switch from one state to the other under external forces. We characterize the required triggering (switching) force with respect to design parameters (e.g., the initial shape of the module). We then apply the soft bistable module to develop fast and strong soft robots, whose triggering forces are generated by a soft actuator - twisted-and-coiled actuator (TCA). We demonstrate a soft gripper that can hold weights more than 8 times its own weight, and a soft jumping robot that can jump more than 5 times its body height. We envision our strategies will overcome the weakness of soft robots to unleash their potential for diverse applications.

We present TrueRMA, a data-efficient, model-free method to learn cost-optimized robot trajectories over a wide range of starting points and endpoints. The key idea is to calculate trajectory waypoints in Cartesian space by recursively predicting orthogonal adaptations relative to the midpoints of straight lines. We generate a differentiable path by adding circular blends around the waypoints, calculate the corresponding joint positions with an inverse kinematics solver and calculate a time-optimal parameterization considering velocity and acceleration limits. During training, the trajectory is executed in a physics simulator and costs are assigned according to a user-specified cost function which is not required to be differentiable. Given a starting point and an endpoint as input, a neural network is trained to predict midpoint adaptations that minimize the cost of the resulting trajectory via reinforcement learning. We successfully train a KUKA iiwa robot to keep a ball on a plate while moving between specified points and compare the performance of TrueRMA against two baselines. The results show that our method requires less training data to learn the task while generating shorter and faster trajectories.

We design and test a 3 DoF parallel cable system capable of applying precise and accurate impulses to walking and running subjects for the University of Utah's Treadport Active Wind Tunnel (TPAWT). Using Nexus VICON motion capture and gait algorithms, perturbations can be applied at different points in the subject's gait. The use of a PID force controller allow the system to create omnidirectional perturbations with walking and running subjects while having the capability to vary amplitude and direction of perturbations. Analysis is presented of the workspace of the large treadmill to test whether the workspace available to activate these perturbations is safe. This paper reports the efficacy of the system and evaluates how warning a runner before impact may affect their displacement. Participants experienced 48 perturbations while running applied with a random combination of a front/back/left/right impact at either toe-off or mid-stance with or without warning. A two sample T-test reveals that warning a runner before impact significantly reduced the magnitude they were displaced for both toe-off (t(46) = 4.98 p<; .001) and mid-stance (t(46) = 3.44, p = .001).

Relatively little work in human and robot control has examined the control of underactuated objects with internal dynamics, such as transporting a cup of coffee, a task that presents little problems for humans. This study examined how humans move a `cup of coffee' with a view to identify principles that may be useful for robot control. The specific focus was on how humans choose initial conditions to safely reach a steady state. We hypothesized that subjects choose initial conditions that minimized the transient duration to reach the steady state faster, as it presented more predictable dynamics. In the experiment, the cup of coffee was reduced to a 2-D cup with a sliding ball inside which was simulated in a virtual environment. Human subjects interacted with this virtual object via a robotic manipulandum that provided haptic feedback. Participants moved the cup between two targets without losing the ball; they were instructed to explore different initial conditions before initiating the continuous interaction. Results showed that subjects converged to a small set of initial conditions that decreased their transient durations and achieved a predictable steady state faster. Simulations with a simple feedforward controller and inverse dynamics calculations confirmed that these initial conditions indeed led to shorter transients and less complex interaction forces. These results may inform robot control of objects with internal dynamics where the effects of initial conditions need further investigation.

While robots can learn models to solve many manipulation tasks from raw visual input, they cannot usually use these models to solve new problems. On the other hand, symbolic planning methods such as STRIPS have long been able to solve new problems given only a domain definition and a symbolic goal, but these approaches often struggle on the real world robotic tasks due to the challenges of grounding these symbols from sensor data in a partially-observable world. We propose Deep Planning Domain Learning (DPDL), an approach that combines the strengths of both methods to learn a hierarchical model. DPDL learns a high-level model which predicts values for a large set of logical predicates consisting of the current symbolic world state, and separately learns a low-level policy which translates symbolic operators into executable actions on the robot. This allows us to perform complex, multistep tasks even when the robot has not been explicitly trained on them. We show our method on manipulation tasks in a photorealistic kitchen scenario.

Grasping in cluttered scenes is challenging for robot vision systems, as detection accuracy can be hindered by partial occlusion of objects. We adopt a reinforcement learning (RL) framework and 3D vision architectures to search for feasible viewpoints for grasping by the use of hand-mounted RGB-D cameras. To overcome the disadvantages of photo-realistic environment simulation, we propose a large-scale dataset called Real Embodied Dataset (RED), which includes full-viewpoint real samples on the upper hemisphere with amodal annotation and enables a simulator that has real visual feedback. Based on this dataset, a practical 3-stage transferable active grasping pipeline is developed, that is adaptive to unseen clutter scenes. In our pipeline, we propose a novel mask-guided reward to overcome the sparse reward issue in grasping and ensure category-irrelevant behavior. The grasping pipeline and its possible variants are evaluated with extensive experiments both in simulation and on a real-world UR-5 robotic arm.

We present a novel class of robots belonging to Constrained Collaborative Mobile Agents (CCMA) family which consists of ground mobile bases with non-holonomic constraints. Moreover, these mobile robots are constrained by closed-loop kinematic chains consisting of revolute joints which can be either passive or actuated. We also describe a novel trajectory optimization method which is general with respect to number of mobile robots, topology of the closed- loop kinematic chains and placement of the actuators at the revolute joints. We also extend the standalone trajectory optimization method to optimize concurrently the design parameters and the control policy. We describe various CCMA system examples, in simulation, differing in design, topology, number of mobile robots and actuation space. The simulation results for standalone trajectory optimization with fixed design parameters is presented for CCMA system examples. We also show how this method can be used for tasks other than end-effector positioning such as internal collision avoidance and external obstacle avoidance. The concurrent design and control policy optimization is demonstrated, in simulations, to increase the CCMA system workspace and manipulation capabilities. Finally, the trajectory optimization method is validated in experiments through two 4-DOF prototypes consisting of 3 tracked mobile bases.

This work presents a method to control omnidirectional micro aerial vehicles (OMAVs) for the tracking of 6-DoF trajectories in free space. A rigid body model based approach is applied in a receding horizon fashion to generate optimal wrench commands that can be constrained to meet limits given by the mechanical design and actuators of the platform. Allocation of optimal actuator commands is performed in a separate step. A disturbance observer estimates forces and torques that may arise from unmodeled dynamics or external disturbances and fuses them into the optimization to achieve offset-free tracking. Experiments on a fully overactuated MAV show the tracking performance and compare it against a classical PD-based controller.

Planning a trajectory with guaranteed safety is a core part for a risk-free flight of a multirotor. If a trajectory planner only aims to ensure safety, it may generate trajectories which overly bypass risky regions and prevent the system from achieving specific missions. This work presents a robust trajectory planning algorithm which simultaneously guarantees the safety and reachability to the target state in the presence of unknown disturbances. We first characterize how the forward and backward reachable sets (FRSs and BRSs) are constructed by using Hamilton-Jacobi reachability analysis. Based on the analysis, we present analytic expressions for the reachable sets and then propose minimal ellipsoids which closely approximate the reachable sets. In the planning process, we optimize the reference trajectory to connect the FRSs and BRSs, while avoiding obstacles. By combining the FRSs and BRSs, we can guarantee that any state inside of the initial set reaches the target set. We validate the proposed algorithm through a simulation of traversing a narrow gap.

This paper presents a trajectory optimization formulation for planning dynamic trajectories of a six-degree-of-freedom (six-DOF) cable-suspended parallel robot (CSPR) that extend beyond the static workspace. The optimization is guided by low-dimensional dynamic models to overcome the local minima and accelerate the exploration of the narrow feasible state space. The dynamic similarity between the six-DOF CSPR and the three-DOF point-mass CSPR is discussed with the analyses of their feasible force polyhedra. Finally, the transition trajectories of a three-DOF CSPR are used as the initial guess of the translational part of the six-DOF motion. With the proposed approach, highly dynamic motions for a six-DOF CSPR are efficiently generated with multiple oscillations. The feasibility is demonstrated by point-to-point and periodic trajectories in the physics simulation.

Correlation filter (CF) has recently exhibited promising performance in visual object tracking for unmanned aerial vehicle (UAV). Such online learning method heavily depends on the quality of the training-set, yet complicated aerial scenarios like occlusion or out of view can reduce its reliability. In this work, a novel time slot-based distillation approach is proposed to efficiently and effectively optimize the training-set's quality on the fly. A cooperative energy minimization function is established to score the historical samples adaptively. To accelerate the scoring process, frames with high confident tracking results are employed as the keyframes to divide the tracking process into multiple time slots. After the establishment of a new slot, the weighted fusion of the previous samples generates one key-sample, in order to reduce the number of samples to be scored. Besides, when the current time slot exceeds the maximum frame number, which can be scored, the sample with the lowest score will be discarded. Consequently, the training-set can be efficiently and reliably distilled. Comprehensive tests on two well-known UAV benchmarks prove the effectiveness of our method with real-time speed on single CPU.

Deep learning has become an increasingly common technique for various control problems, such as robotic arm manipulation, robot navigation, and autonomous vehicles. However, the downside of using deep neural networks to learn control policies is their opaque nature and the difficulties of validating their safety. As the networks used to obtain state-of-the-art results become increasingly deep and complex, the rules they have learned and how they operate become more challenging to understand. This presents an issue, since in safety-critical applications the safety of the control policy must be ensured to a high confidence level. In this paper, we propose an automated black box testing framework based on adversarial reinforcement learning. The technique uses an adversarial agent, whose goal is to degrade the performance of the target model under test. We test the approach on an autonomous vehicle problem, by training an adversarial reinforcement learning agent, which aims to cause a deep neural network-driven autonomous vehicle to collide. Two neural networks trained for autonomous driving are compared, and the results from the testing are used to compare the robustness of their learned control policies. We show that the proposed framework is able to find weaknesses in both control policies that were not evident during online testing and therefore, demonstrate a significant benefit over manual testing methods.

The best way to combine the results of deep learning with standard 3D reconstruction pipelines remains an open problem. While systems that pass the output of traditional multi-view stereo approaches to a network for regularisation or refinement currently seem to get the best results, it may be preferable to treat deep neural networks as separate components whose results can be probabilistically fused into geometry- based systems. Unfortunately, the error models required to do this type of fusion are not well understood, with many different approaches being put forward. Recently, a few systems have achieved good results by having their networks predict probability distributions rather than single values. We propose using this approach to fuse a learned single-view depth prior into a standard 3D reconstruction system. Our system is capable of incrementally producing dense depth maps for a set of keyframes. We train a deep neural network to predict discrete, nonparametric probability distributions for the depth of each pixel from a single image. We then fuse this "probability volume" with another probability volume based on the photometric consistency between subsequent frames and the keyframe image. We argue that combining the probability volumes from these two sources will result in a volume that is better conditioned. To extract depth maps from the volume, we minimise a cost function that includes a regularisation term based on network predicted surface normals and occlusion boundaries. Through a series of experiments, we demonstrate that each of these components improves the overall performance of the system.

Photogrammetry techniques used for 3D reconstructions and motion estimation from images are based on projective geometry that models the image formation process. However, in the underwater setting, refraction of light rays at the housing interface introduce non-linear effects in the image formation. These effects produce systematic errors if not accounted for, and severely degrade the quality of the acquired images. In this paper, we present a novel approach to the calibration of cameras inside spherical domes with large offsets between dome and camera centers. Such large offsets not only amplify the effect of refraction, but also introduce blur in the image that corrupts feature extractors used to establish image-world correspondences in existing refractive calibration methods. We propose using the point spread function (PSF) as a complete description of the optical system and introduce a procedure to recover the camera pose inside the dome based on the measurement of the distortions. Results on a collected dataset show the method is capable of recovering the camera pose with high accuracy.

The flight characteristics of bats remarkably have been overlooked in aerial drone designs. Unlike other animals, bats leverage the manipulation of inertial dynamics to exhibit aerial flip turns when they perch. Inspired by this unique maneuver, this work develops and uses a tiny robot called Harpoon to demonstrate that the preparation for upside-down landing is possible through: 1) reorientation towards the landing surface through zero-angular-momentum turns and 2) reaching to the surface through shooting a detachable landing gear. The closed-loop manipulations of inertial dynamics takes place based on a symplectic description of the dynamical system (body and appendage), which is known to exhibit an excellent geometric conservation properties.

In this paper, a novel human-robot collaborative framework for mixed case palletizing is presented. The framework addresses several challenges associated with the detection and localisation of boxes and pallets through visual perception algorithms, high-level optimisation of the collaborative effort through effective role-allocation principles, and maximisation of packing density. A graphical user interface (GUI) is additionally developed to ensure an intuitive allocation of roles and the optimal placement of the boxes on target pallets. The framework is evaluated in two conditions where humans operate with and without the support of a Mobile COllaborative robotic Assistant (MOCA). The results show that the optimised placement can improve up to the 20% with respect to a manual execution of the same task, and reveal the high potential of MOCA in increasing the performance of collaborative palletizing tasks.

Safety in Human-Robot Collaboration (HRC) is a bottleneck to HRC-productivity in industry. With robots being the main source of hazards, safety engineers use over-emphasized safety measures, and carry out lengthy and expensive risk assessment processes on each HRC-layout reconfiguration. Recent advances in deep Reinforcement Learning (RL) offer solutions to add intelligence and comprehensibility of the environment to robots. In this paper, we propose a framework that uses deep RL as an enabling technology to enhance intelligence and safety of the robots in HRC scenarios and, thus, reduce hazards incurred by the robots. The framework offers a systematic methodology to encode the task and safety requirements and context of applicability into RL settings. The framework also considers core components, such as behavior explainer and verifier, which aim for transferring learned behaviors from research labs to industry. In the evaluations, the proposed framework shows the capability of deep RL agents learning collision-free point-to-point motion on different robots inside simulation, as shown in the supplementary video.

Developing autonomous vehicles capable of navigating safely and socially around pedestrians is a major challenge in intelligent transportation. This challenge cannot be met without understanding pedestrians' behavioral response to an autonomous vehicle, and the task of building a clear and quantitative description of the pedestrian to vehicle interaction remains a key milestone in autonomous navigation research. As a step towards safe proactive navigation in a space shared with pedestrians, this work introduces a pedestrian-vehicle interaction behavioral model. The model estimates the pedestrian's cooperation with the vehicle in an interaction scenario by a quantitative time-varying function. Using this cooperation estimation the pedestrian's trajectory is predicted by a cooperation-based trajectory planning model. Both parts of the model are tested and validated using real-life recorded scenarios of pedestrian-vehicle interaction. The model is capable of describing and predicting agents' behaviors when interacting with a vehicle in both lateral and frontal crossing scenarios.