ICRA 2019 论文列表

We explore the characteristics of secondary contacts when applying forces with the end-effector of a robot and address the question when these secondary contacts can increase maximum applicable end-effector forces or reduce required actuator efforts. To this end, we formalize the effect of such secondary contacts in terms of required actuator efforts and derive efficiency bounds depending on the contact characteristics and robot configuration. Our findings are confirmed by experiments with a redundant serial manipulator.

Robotic grasping can enable mobile vehicles to physically interact with the environment for delivery, repositioning, or landing. However, the requirements for grippers on mobile vehicles differ substantially from those used for conventional manipulation. Specifically, grippers for dynamic mobile robots should be capable of rapid activation, high force density, low power consumption, and minimal computation. In this work, we present a biologically-inspired robotic gripper designed specifically for mobile platforms. This design exploits a bistable shell to achieve “reflexive” activation based on contact with the environment. The mechanism can close its grasp within 0. 12s without any sensing or control. Electrical input power is not required for grasping or holding load. The reflexive gripper utilizes a novel pneumatic design to open its grasp with low power, and the gripper can carry slung loads up to 28 times its weight. This new mechanism, including the kinematics, static behavior, control structure, and fabrication, is described in detail. A proof of concept prototype is designed, built, and tested. Experimental results are used to characterize performance and demonstrate the potential of these methods.

Humans commonly exploit interaction with the environment constraints to assist the execution of the loco-manipulation tasks they perform. One particular example is the exploration of contacts during manipulation to relax the loading of those arm joints that are not directly involved in the generation of the manipulation motions and forces, e.g. establishing a contact with the elbow joint to reduce the effort of the upper arm while executing wrist level manipulation. In this paper, we shall explore the possibility of actively (a) utilizing the environment for a non-end-effector support contact towards reducing the joints efforts during manipulation tasks. This is achieved by our proposed control scheme with a three-level hierarchical compliance controller. The highest priority task is assigned to an impedance control that regulates the interaction at the contact control point on the arm in the normal direction of the support plane prior to contact, and is switched to an optimal contact force control for minimizing the joint effort after the contact is built. The second priority task is an impedance control at the same point in the tangential directions of the plane to stabilize the contact. In the end, an impedance behavior at the end-effector is designed to deal with the interaction forces required by the manipulation tasks. The efficacy of the proposed control scheme was corroborated by simulations and experiments, where significant joint effort reduction was observed.

Humanoid robots typically display locomotion patterns that include walking with flat foot-ground contact, and knees slightly bent. However, analysis of human gait indicate that several physiological mechanisms like stretched knees, heel-strike and toe push-off increase the step length and energetic efficiency of locomotion. This paper presents an implementation of two of those mechanisms, namely stretched knees and push-off, on a quasi-static whole-body balancing controller. The influence of such mechanisms on the kinematic capabilities of the DLR humanoid robot TORO is analyzed in different experiments, and their benefits are thoroughly discussed. As a result, the energetic savings of balancing with stretched knees are shown to be of reduced magnitude with respect to the overall power consumption of the robot, and the ability of TORO for negotiating stairs is greatly enhanced.

Biped robots require substantial amounts of power alternately on each leg while walking, hopping, and running. However, it is difficult to mount high-power large electrical motors in conventional mechanical transmission systems owing to spatial limitations. A hydraulic direct-drive system is proposed in which the size of the motor in each leg can be reduced by sharing the motor outputs between the legs. In this paper, the hydraulic direct-drive system is evaluated in an actual hydraulic system. Velocity followability, excellent energy saving, and virtually perfect position tracking are achieved with the proposed system. The results of performance comparison with a valve-based control system show that energy consumption is controlled and good position-following capability is achieved using the proposed system.

In this work, we present an application of a policy gradient algorithm to a real-time robotic learning problem, where the goal is to maximize the average lift generation of a dynamically scaled robotic wing at a constant Reynolds number (Re). Compared to our previous work, the merit of this work is two-fold. First, a central pattern generator (CPG) model was used as the motion controller, which provided a smooth generation and transition of rhythmic wing motion patterns while the CPG was being updated by the policy gradient, thereby accelerating the sample generation and reducing the total learning time. Second, the kinematics included three degrees of freedom (stroke, deviation, pitching) and were also free of half-stroke symmetry constraint, together they yielded a larger kinematic space which later explored by the policy gradient to maximize the lift generation. The learned wing kinematics used the full range of stroke and deviation to maximize the lift generation, implying that the wing trajectories with larger disk area and lower frequencies were preferred for high lift generation at constant Re. Furthermore, the wing pitching amplitude converged to values between 45°-49° regardless of what the other parameters were. Notably, the learning agent was able to find two locally optimal wing motion patterns, which had distinct shapes of wing trajectory but generated similar cycle-averaged lift.

On-orbit servicing of satellites is complicated by the fact that almost all existing satellites were not designed to be serviced. This creates a number of challenges, one of which is to cut and partially remove the protective thermal blanketing that encases a satellite prior to performing the servicing operation. A human operator on Earth can perform this task telerobotically, but must overcome difficulties presented by the multi-second round-trip telemetry delay between the satellite and the operator and the limited, or even obstructed, views from the available cameras.This paper reports the results of ground-based experiments with trained NASA robot teleoperators to compare our recently-reported augmented virtuality visualization to the conventional camera-based visualization. We also compare the master console of a da Vinci surgical robot to the conventional teleoperation interface. The results show that, for the cutting task, the augmented virtuality visualization can improve operator performance compared to the conventional visualization, but that operators are more proficient with the conventional control interface than with the da Vinci master console.

This paper presents the first practical demonstration of a recently developed theory of balance control that aims to achieve high performance in the sense of allowing a robot to make large, fast movements while maintaining its balance on a narrow support. This theory includes a simple method of leaning in anticipation of future motion commands, which is largely responsible for the high performance. The experiments reported here use a robot acting as a reaction wheel pendulum, and they test only the 2-D version of the theory. The results show that the balance controller's performance in practice closely resembles its theoretical performance. This paper also presents a simple yet accurate balance offset observer that measures the difference between true and estimated balanced configurations.

This paper presents experiments to assess the plume mapping performance of autonomous robots. The paper compares several mapping algorithms including Gaussian Process regression, Neural networks and polynomial and piecewise linear interpolation. The methods are compared in Monte Carlo simulations using a well known plume model and in indoor experiments using a ground robot. Unlike previous work on mapping using unmanned vehicles, the indoor experiments were performed in a controlled and repeatable manner where a steady state ground truth could be obtained in order to properly assess the various regression methods using data from a real dispersive source and sensor. The effect of sampling time during data collection was assessed with regards to the mapping accuracy, and the data collected during the experiments have been made available. Overall, the Gaussian Process method was found to perform the best among the regression algorithms, showing more robustness to the noisy measurements obtained from short sampling periods, enabling an accurate map to be produced in significantly less time. Finally, plume mapping results are presented in uncontrolled outdoor conditions, using an unmanned aerial vehicle, to demonstrate the system in a realistic uncontrolled environment.

Non-Gaussian and multimodal distributions are an important part of many recent robust sensor fusion algorithms. In difference to robust cost functions, they are probabilistically founded and have good convergence properties. Since their robustness depends on a close approximation of the real error distribution, their parametrization is crucial. We propose a novel approach that allows to adapt a multi-modal Gaussian mixture model to the error distribution of a sensor fusion problem. By combining expectation-maximization and non-linear least squares optimization, we are able to provide a computationally efficient solution with well-behaved convergence properties. We demonstrate the performance of these algorithms on several real-world GNSS and indoor localization datasets. The proposed adaptive mixture algorithm outperforms state-of-the-art approaches with static parametrization. Source code and datasets are available under https://mytuc.org/libRSF.

Since the spread of robotic systems embedding in their mechanics purposefully designed elastic elements, the interest in characterizing and exploiting non-linear oscillatory behaviors has progressively grown. However, few works so far looked at the problem from the point of view of modal analysis. This is particularly surprising if considered the central role that modal theory had in the development of classic results in analysis and control of linear mechanical systems. With the aim of making a step toward translating and extending this powerful tool to the robotic field, we present the complete modal characterization of a simple yet representative non-linear elastic robot: the 2D planar mass-spring-damper system. Generic non-linear elastic forces and dissipative effects are considered. We provide here exact descriptions of the two non-linear normal modes of the system. We then extend the analysis to generic combinations of the modes in conservative case and for small damping. Simulations are provided to illustrate the theoretical results. This is one of the very firsts applications of normal mode theory to dynamically coupled non-linear systems, and the first exact result in the field.

Practical use of robots in diverse domains requires programming for, or adapting to, each domain and its unique characteristics. Failure to do so compromises the ability of the robot to achieve task-relevant objectives. Here we describe how the learned terrain reaction force profiles of a hopping robot serve the additional objectives of classifying terrain and quickly learning control strategies to accomplish a jumping task on novel terrain. We show that the reaction forces experienced during closed-loop jumping are sufficient to discriminate between three different terrain types (granular, trampoline, and rigid) when using the learned models as discriminators. Building on this, we show that applying the classification to unknown terrain types leads to faster task completion, where the task objective is to meet a specific jump height. The classification experiments, utilizing real-world jumping data, achieve 95% prediction accuracy. The online learning experiments leverage simulation as there is more control over the terrain properties. Terrain-informed learning achieves the target hop heights more than 2x faster than without terrain knowledge when the prediction is correct, and 1.5x faster when the prediction is incorrect. Thus, applying the closest approximately known terrain knowledge facilitates low shot learning when hopping on unknown terrain.

Event cameras are novel bio-inspired vision sensors that output pixel-level intensity changes, called “events”, instead of traditional video images. These asynchronous sensors naturally respond to motion in the scene with very low latency (microseconds) and have a very high dynamic range. These features, along with a very low power consumption, make event cameras an ideal sensor for fast robot localization and wearable applications, such as AR/VR and gaming. Considering these applications, we present a method to track the 6-DOF pose of an event camera in a known environment, which we contemplate to be described by a photometric 3D map (i.e., intensity plus depth information) built via classic dense 3D reconstruction algorithms. Our approach uses the raw events, directly, without intermediate features, within a maximum-likelihood framework to estimate the camera motion that best explains the events via a generative model. We successfully evaluate the method using both simulated and real data, and show improved results over the state of the art. We release the datasets to the public to foster reproducibility and research in this topic.

In this work, we present an information based exploration strategy tailored for the generation of high resolution 3D maps. We employ RGBD panoramas because they have been shown to provide memory efficient high quality representations of space. Robots explore the environment by selecting locations with maximal Cauchy-Schwarz Quadratic Mutual Information (CSQMI) computed on an angle enhanced occupancy grid to collect these RGBD panoramas. By employing the angle enhanced occupancy grid, the resulting exploration strategy emphasizes perspective in addition to binary coverage. Furthermore, the goal selection strategy is improved by using image morphology to reduce the search space over which CSQMI is computed. We present experimental results demonstrating the improved performance in perception related tasks by capturing panoramas using this approach, near frontier exploration, and a control of logging images at regular intervals while teleoperating the robot through the workspace. Collect imagery was passed through an object detection library with our perspective aware approach yielding a greater number of successful detections compared to near frontier exploration.

There has been a recent emergence of sampling-based techniques for estimating epistemic uncertainty in deep neural networks. While these methods can be applied to classification or semantic segmentation tasks by simply averaging samples, this is not the case for object detection, where detection sample bounding boxes must be accurately associated and merged. A weak merging strategy can significantly degrade the performance of the detector and yield an unreliable uncertainty measure. This paper provides the first in-depth investigation of the effect of different association and merging strategies. We compare different combinations of three spatial and two semantic affinity measures with four clustering methods for MC Dropout with a Single Shot Multi-Box Detector. Our results show that the correct choice of affinity-clustering combination can greatly improve the effectiveness of the classification and spatial uncertainty estimation and the resulting object detection performance. We base our evaluation on a new mix of datasets that emulate near open-set conditions (semantically similar unknown classes), distant open-set conditions (semantically dissimilar unknown classes) and the common closed-set conditions (only known classes).

The application of robots in inspection tasks has been growing quickly thanks to the advancements in autonomous navigation technology, especially the robot localization techniques in GPS-denied environments. Although many methods have been proposed to localize a robot using onboard sensors such as cameras and LiDARs, achieving robust localization in geometrically degenerated environments, e.g. tunnels, remains a challenging problem. In this work, we focus on the robust localization problem in such situations. A novel degeneration characterization model is presented to estimate the localizability at a given location in the prior map. And the localizability of a LiDAR and an Ultra-Wideband (UWB) ranging radio is analyzed. Additionally, a probabilistic sensor fusion method is developed to combine IMU, LiDAR and the UWB. Experiment results show that this method allows for robust localization inside a long straight tunnel.

Mechanics-based models of thin elastic structures are prevalent in robotics research, both in soft/continuum robot modeling, and in robotic manipulation of strings, sutures, needles, and endoscopes. In all these applications, distributed loads along the device's length can affect its shape in space. Estimation of the distributed loading based on observation of the object's shape constitutes a classical mechanics inverse problem that would be useful in many applications, but this problem has received relatively little attention to date. In this paper, we propose methods to estimate distributed loads on an elastic rod using a large-deflection Cosserat-rod model and constrained nonlinear optimization. We perform experiments that illustrate the feasibility of using these methods to locate regions of high contact force along the rod, and to estimate magnitudes of the forces that are applied. Results show that overall force magnitudes and locations can be estimated with average error of 0.29 N (6.7% of average resultant magnitude) and 4 mm (2% of rod length) for complex double-bend shapes, and the shape approximation has near-zero error.

Most state-of-the-art underwater vision systems are calibrated manually in shallow water and used in open seas without changing. However, the refractivity of the water is adaptively changed depending on the salinity, temperature, depth or other underwater environmental indexes, which inevitably generate the calibration errors and induces incorrectness e.g., for underwater Simultaneously Localization and Mapping (SLAM). To address this issue, in this paper, we propose a new underwater Camera-Inertial Measurement Unit (IMU) calibration model, which just needs to be calibrated once in the air, and then both the intrinsic parameters and extrinsic parameters between the camera and IMU could be automatically calculated depending on the environment indexes. To our best knowledge, this is the first work to consider the underwater Camera-IMU calibration via environmental indexes. We also build a verification platform to validate the effectiveness of our proposed method on real experiments, and use it for underwater monocular Visual-Inertial SLAM.

Passivity has been the most often used constraint for the stable controller design of bilateral teleoperation systems. Especially, Time Domain Passivity Approach (TDPA) has been used in many applications since it has been known as one of the least conservative passivity-based approaches. Although TDPA were able to stabilize the system with the least conservatism, it has its own drawbacks as the cost of achieving the least conservative passivity especially when there is communication time-delay. Due to the on/off bang-bang control-like modification for instantaneous passivity recovery, it has high frequency force vibrations on the slave and especially master side. By implementing a virtual mass-spring system between the passivity controller and master device, these high frequency vibration has been eliminated. However, the gains need proper tuning as they are dependent on the teleoperation setup and application. It also tends to make the system sluggish which further distorts the transparency. We propose a new observer-based gradient controller to eliminate the force jittering on the master side. It rectifies the delayed feedback force by removing the undesired increase in force which is generated by the delay in communication channel. It does not require any system parameters and there are no gains to tune, thus it can be added to any teleoperator irrespective of its dynamics and without having any prior system information. The proposed approach was implemented to a position-force bilateral teleoperation system, and compared with TDPA with virtual mass spring with round-trip delays of up to 500 ms for hard wall contacts.