ICRA 2019 论文列表

We present structured domain randomization (SDR), a variant of domain randomization (DR) that takes into account the structure of the scene in order to add context to the generated data. In contrast to DR, which places objects and distractors randomly according to a uniform probability distribution, SDR places objects and distractors randomly according to probability distributions that arise from the specific problem at hand. In this manner, SDR-generated imagery enables the neural network to take the context around an object into consideration during detection. We demonstrate the power of SDR for the problem of 2D bounding box car detection, achieving competitive results on real data after training only on synthetic data. On the KITTI easy, moderate, and hard tasks, we show that SDR outperforms other approaches to generating synthetic data (VKITTI, Sim 200k, or DR), as well as real data collected in a different domain (BDD100K). Moreover, synthetic SDR data combined with real KITTI data outperforms real KITTI data alone.

Motion planning for underwater vehicles must consider the effect of ocean currents. We present an efficient method to compute reachability and cost between sample points in sampling-based motion planning that supports long-range planning over hundreds of kilometres in complicated flows. The idea is to search a reduced space of control inputs that consists of stream functions whose level sets, or streamlines, optimally connect two given points. Such stream functions are generated by superimposing a control input onto the underlying current flow. A streamline represents the resulting path that a vehicle would follow as it is carried along by the current given that control input. We provide rigorous analysis that shows how our method avoids exhaustive search of the control space, and demonstrate simulated examples in complicated flows including a traversal along the east coast of Australia, using actual current predictions, between Sydney and Brisbane.

This paper proposes a bandwidth tunable technique for real-time probabilistic scene modeling and mapping to enable co-robotic exploration in communication constrained environments such as the deep sea. The parameters of the system enable the user to characterize the scene complexity represented by the map, which in turn determines the bandwidth requirements. The approach is demonstrated using an underwater robot that learns an unsupervised scene model of the environment and then uses this scene model to communicate the spatial distribution of various high-level semantic scene constructs to a human operator. Preliminary experiments in an artificially constructed tank environment as well as simulated missions over a 10m×10m coral reef using real data show the tunability of the maps to different bandwidth constraints and science interests. To our knowledge this is the first paper to quantity how the free parameters of the unsupervised scene model impact both the scientific utility of and bandwidth required to communicate the resulting scene model.

Robotic arms built from stiffness-adjustable, continuously bending segments serially connected with revolute joints have the ability to change their mechanical architecture and workspace, thus allowing high flexibility and adaptation to different tasks with less than six degrees of freedom, a concept that we call malleable robots. Known stiffening mechanisms may be used to implement suitable links for these novel robotic manipulators; however, these solutions usually show a reduced performance when bending due to structural deformation. By including an inner support structure this deformation can be minimised, resulting in an increased stiffening performance. This paper presents a new multi-material spine-inspired flexible structure for providing support in stiffness-controllable layer-jamming-based robotic links of large diameter. The proposed spine mechanism is highly movable with type and range of motions that match those of a robotic link using solely layer jamming, whilst maintaining a hollow and light structure. The mechanics and design of the flexible spine are explored, and a prototype of a link utilising it is developed and compared with limb segments based on granular jamming and layer jamming without support structure. Results of experiments verify the advantages of the proposed design, demonstrating that it maintains a constant central diameter across bending angles and presents an improvement of more than 203% of resisting force at 180°.

This paper presents a novel continuum robot sheath for use in single-port minimally invasive procedures such as neuroendoscopy in which the sheath is designed to deliver multiple robotic arms. Articulation of the sheath is achieved by using precurved superelastic tubes lining the working channels used for arm delivery. These tubes perform a similar role to push/pull tendons, but can accomplish shape change of the sheath via rotation as well as translation. A kinematic model using Cosserat rod theory is derived which is based on modeling the system as a set of eccentrically aligned precurved tubes constrained along their length by an elastic backbone. The specific case of a two-arm sheath is considered in detail and its relationship to a concentric tube balanced pair is described. Simulation and experiment are used to investigate the concept, map its workspace and to evaluate the kinematic model.

This paper reports the development of a compact optical-acoustic frontend module for underwater communication and ranging. The module is enabled by a new water-immersible MEMS scanning mirror (WIMSM). It is capable of transmitting, receiving and steering co-centered and co-directional laser and ultrasound beams under water. To monitor its rotating angle in real time, scan position sensors based on Hall effect have been integrated into the WIMSM. The angular alignment of the laser and ultrasound beams in both transmission and reception modes has been examined. The experimental results show that the laser and ultrasound beams can remain aligned with less than 2.1 degrees under envelope of pan and tilt rotations. This capability is critical for the continuing development of the new bi-modal communication and ranging underwater Vehicles (AUVs).

This paper introduces a Bayesian state estimator for contact-rich manipulation tasks with application in non-prehensile manipulation, industrial assembly or in-hand localization. The core idea of our approach is to explicitly model both the contact dynamics and a torque-based robot controller as part of the underlying system model. Our approach is capable of estimating the state of movable objects for various robot kinematics and geometries of robots and objects. This includes complex scenarios with multiple robots, multiple objects and articulated objects. We have validated our approach in simulation and on a physical robot. The experiments show that multimodal distributions of six degrees of freedom object poses can be accurately tracked in real-time in a complex manipulation scenario.

This paper presents Stanford Doggo, a quasi-direct-drive quadruped capable of dynamic locomotion. This robot matches or exceeds common performance metrics of state-of-the-art legged robots. In terms of vertical jumping agility, a measure of average vertical speed, Stanford Doggo matches the best performing animal and surpasses the previous best robot by 22%. An overall design architecture is presented with focus on our quasi-direct-drive design methodology. The hardware and software to replicate this robot is open-source, requires only hand tools for manufacturing and assembly, and costs less than $3000.

We consider dynamic stair climbing with the HRP-4 humanoid robot as part of an Airbus manufacturing use-case demonstrator. We share experimental knowledge gathered so as to achieve this task, which HRP-4 had never been challenged to before. In particular, we extend walking stabilization based on linear inverted pendulum tracking [1] by quadratic programming-based wrench distribution and a whole-body admittance controller that applies both end-effector and CoM strategies. While existing stabilizers tend to use either one or the other, our experience suggests that the combination of these two approaches improves tracking performance. We demonstrate this solution in an on-site experiment where HRP4 climbs an industrial staircase with 18.5 cm high steps, and release our walking controller as open source software.

Recent progress in the field of robotic manipulation has generated interest in fully automatic object packing in warehouses. This paper proposes a formulation of the packing problem that is tailored to the automated warehousing domain. Besides minimizing waste space inside a container, the problem requires stability of the object pile during packing and the feasibility of the robot motion executing the placement plans. To address this problem, a set of constraints are formulated, and a constructive packing pipeline is proposed to solve these constraints. The pipeline is able to pack geometrically complex, non-convex objects while satisfying stability and robot packability constraints. In particular, a new 3D positioning heuristic called Heightmap-Minimization heuristic is proposed, and heightmaps are used to speed up the search. Experimental evaluation of the method is conducted with a realistic physical simulator on a dataset of scanned real-world items, demonstrating stable and high-quality packing plans compared with other 3D packing methods.

In this paper, we present a constrained optimization framework for evaluating the post-contact stability of underactuated precision grasping configurations with a single degree of actuation. Relationships between key anthropomorphic design parameters including link length ratios, transmission ratios, joint stiffness ratios and palm width are developed with applications in upper limb prosthetic design. In addition to grasp stability, we examine post-contact system work, to reduce reconfiguration, and consider the range of objects that can be stably grasped. External wrenches were simulated on a subset of the heuristically evaluated optimal solutions and an optimal configuration was experimentally tested to determine favorable wrench resistible gripper orientations for grasp planning applications.

Earlier work demonstrates the promise of deep-learning-based approaches for point cloud segmentation; however, these approaches need to be improved to be practically useful. To this end, we introduce a new model SqueezeSegV2. With an improved model structure, SqueezeSetV2 is more robust against dropout noises in LiDAR point cloud and therefore achieves significant accuracy improvement. Training models for point cloud segmentation requires large amounts of labeled data, which is expensive to obtain. To sidestep the cost of data collection and annotation, simulators such as GTA-V can be used to create unlimited amounts of labeled, synthetic data. However, due to domain shift, models trained on synthetic data often do not generalize well to the real world. Existing domain-adaptation methods mainly focus on images and most of them cannot be directly applied to point clouds. We address this problem with a domain-adaptation training pipeline consisting of three major components: 1) learned intensity rendering, 2) geodesic correlation alignment, and 3) progressive domain calibration. When trained on real data, our new model exhibits segmentation accuracy improvements of 6.0-8.6% over the original SqueezeSeg. When training our new model on synthetic data using the proposed domain adaptation pipeline, we nearly double test accuracy on real-world data, from 29.0% to 57.4%. Our source code and synthetic dataset are open sourced11https://github.com/xuanyuzhou98/SqueezeSegV2

This paper investigates the curve path following problem for a class of planar underactuated bio-inspired snake robots. The time-varying line-of-sight (LOS) guidance law and the cubic spline interpolation (CSI) path-planning method are employed. Existing studies focus on straight line path following which only gives a solution for snake robot motion control in relatively simple environments. Considering the snake robot's many degrees of freedom and excellent mobility in terrains, we propose a more applicable solution of curve path following for snake robots on the ground. The improved LOS helps the snake robot to steer aggressively at a sharp turning point. Furthermore, to avoid the sideslip of the snake robot caused by the ground friction change, an integral controller is introduced in the design of the heading reference. Simulations and experiments on an 8-link custom-built snake robot are conducted and the results demonstrate and validate the effectiveness of the proposed curve path following algorithm.

Sensors producing 3D point clouds such as 3D laser scanners and RGB-D cameras are widely used in robotics, be it for autonomous driving or manipulation. Aligning point clouds produced by these sensors is a vital component in such applications to perform tasks such as model registration, pose estimation, and SLAM. Iterative closest point (ICP) is the most widely used method for this task, due to its simplicity and efficiency. In this paper we propose a novel method which solves the optimisation problem posed by ICP using stochastic gradient descent (SGD). Using SGD allows us to improve the convergence speed of ICP without sacrificing solution quality. Experiments using Kinect as well as Velodyne data show that, our proposed method is faster than existing methods, while obtaining solutions comparable to standard ICP. An additional benefit is robustness to parameters when processing data from different sensors.

Tracking of human legs during walking are key technologies for gait analysis evaluating the movement function of the elderly and patients with gait disorders. Although the motion capture cameras are the gold standard method for gait analysis because of their high accuracy, they are not always accessible in clinical sites because of their cost, scale, and usability. In response, a laser range sensor (LRS), which is used for obstacle avoidance and human detection of mobile robots, has recently been employed for tracking of leg motions. Some previous studies set LRS at shin height and tracked leg motions during walking using three or five observation patterns and the Kalman filtering and data association methods. However, these systems had difficulty in tracking during walking along a circular trajectory including frequent overlaps and occlusions of legs. Therefore, this paper presents a spatiotemporal and kinetic gait analysis system using a single LRS and instrumented insoles and proposes a multisensor fusion algorithm for tracking leg motions. The instrumented insoles are in-shoe devices embedded force sensors and can detect accurate timings of gait events via force sensing. The system identifies gait phases by the fusion algorithm and switches acceleration input added to motion models of tracked legs for the Kalman filter and data association. The tracking performance of the proposed system was evaluated by measuring walking on a circular trajectory in experiments.

We propose an efficient spatio-temporal model for mobile autonomous robots operating in human populated environments. Our method aims to model periodic temporal patterns of people presence, which are based on peoples' routines and habits. The core idea is to project the time onto a set of wrapped dimensions that represent the periodicities of people presence. Extending a 2D spatial model with this multidimensional representation of time results in a memory efficient spatio-temporal model. This model is capable of long-term predictions of human presence, allowing mobile robots to schedule their services better and to plan their paths. The experimental evaluation, performed over datasets gathered by a robot over a period of several weeks, indicates that the proposed method achieves more accurate predictions than the previous state of the art used in robotics.

Detection of spatial change around a robot is indispensable in several robotic applications, such as search and rescue, security, and surveillance. The present paper proposes a fast spatial change detection technique for a mobile robot using an on-board RGB-D/stereo camera and a highly precise 3D map created by a 3D laser scanner. This technique first converts point clouds in a map and measured data to grid data (ND voxels) using normal distributions transform and classifies the ND voxels into three categories. The voxels in the map and the measured data are then compared according to the category and features of the ND voxels. Overlapping and voting techniques are also introduced in order to detect the spatial changes more robustly. We conducted experiments using a mobile robot equipped with real-time range sensors to confirm the performance of the proposed real-time localization and spatial change detection techniques in indoor and outdoor environments.

We study the problem of controlling a swarm of anonymous, mobile robots to cooperatively cover an unknown two-dimensional space. The novelty of our proposed solution is that it is applicable to extremely simple robots that lack run-time computation or storage. The solution requires only a single bit of information per robot-whether or not another robot is present in its line of sight. Computer simulations show that our deterministic controller, which was obtained through off-line optimization, achieves around 71-76% coverage in a test scenario with no robot redundancy, which corresponds to a 26-39% reduction of the area that is not covered, when compared to an optimized random walk. A moderately lower level of performance was observed in 20 experimental trials with 25 physical e-puck robots. Moreover, we demonstrate that the same controller can be used in environments of different dimensions and even to navigate a maze. The controller provides a baseline against which one can quantify the performance improvements that more advanced and expensive techniques may offer. Moreover, due to its simplicity, it could potentially be implemented on swarms of sub-millimeter-sized robots. This would pave the way for new applications in micro-medicine.

This paper introduces SpaceBok, a quadrupedal robot created to investigate dynamic legged locomotion for the exploration of low-gravity celestial bodies. With a hip height of 500 mm and a mass of 20 kg, its dimensions are comparable to a medium-sized dog. The robot's leg configuration is based on an optimized parallel motion mechanism that allows the integration of parallel elastic elements to store and release energy for powerful jumping maneuvers. High-torque brushless motors in combination with customized single-stage planetary gear transmissions enable force control at the foot contact points based on motor currents. We present successful walking, trotting, and pronking experiments. Thereby, Spacebok achieved maximal jump heights in single jump experiments of up to 1.05 m (more than twice the hip height) and a walking velocity of 1m/s. Moreover, simulation results for low gravity on the moon suggest that our robot can move with up to 1.1m/s at an approximate cost of transport of 1 in moon gravity when using the pronking gait.