ICRA 2019 论文列表

This paper considers a group of mobile sensing agents in a flow field and presents a distributed method for motion tomography (MT) that estimates the underlying flow field. MT formulates an underdetermined nonlinear system of equations as an inverse problem. Inspired by the Kaczmarz method which is an optimization approach for solving a linear system of equations, our previous work developed a nonlinear Kaczmarz method that solves the system of equations associated with MT. Considering distributed multi-agent systems for MT, this paper extends the nonlinear Kaczmarz method into a distributed framework. The distributed nonlinear Kaczmarz method is developed by formulating a constrained consensus problem that belongs to a class of projected consensus algorithms. To study the convergence and consensus for the method, its linear case is analyzed first and then its nonlinear case is discussed. The nonlinear case of the method is further validated through simulations by estimating a gyre flow field using mobile sensor networks with different numbers of neighboring agents. Resulting estimated flow fields are compared with a flow field estimated by its centralized counterpart.

We present a robotic pen-drawing system that is capable of faithfully reproducing pen art on an unknown surface. Our robotic system relies on an industrial, seven-degree-of freedom manipulator that can be both position- and impedance-controlled. In order to estimate a rough geometry of the target, continuous surface, we first generate a point cloud of the surface using an RGB-D camera, which is filtered to remove outliers and calibrated to the physical canvas surface. Then, our control algorithm physically reproduces digital drawing on the surface by impedance-controlling the manipulator. Our impedance-controlled drawing algorithm compensates for the uncertainty and incompleteness inherent to a point-cloud estimation of the drawing surface. Moreover, since drawing 2D vector pen art on a 3D surface requires surface parameterization that does not destroy the original 2D drawing, we rely on the least squares conformal mapping. Specifically, the conformal map reduces angle distortion during surface parameterization. As a result, our system can create distortion-free and complicated pen drawings on general surfaces with many unpredictable bumps robustly and faithfully.

For autonomous vehicles to be able to operate successfully they need to be aware of other vehicles with sufficient time to make safe, stable plans. Given the possible closing speeds between two vehicles, this necessitates the ability to accurately detect distant vehicles. Many current image-based object detectors using convolutional neural networks exhibit excellent performance on existing datasets such as KITTI. However, the performance of these networks falls when detecting small (distant) objects. We demonstrate that incorporating radar data can boost performance in these difficult situations. We also introduce an efficient automated method for training data generation using cameras of different focal lengths.

Despite the recent active research on processing point clouds with deep networks, few attention has been on the sensitivity of the networks to rotations. In this paper, we propose a deep learning architecture that achieves discrete SO(2)/SO(3) rotation equivariance for point cloud recognition. Specifically, the rotation of an input point cloud with elements of a rotation group is similar to shuffling the feature vectors generated by our approach. The equivariance is easily reduced to invariance by eliminating the permutation with operations such as maximum or average. Our method can be directly applied to any existing point cloud based networks, resulting in significant improvements in their performance for rotated inputs. We show state-of-the-art results in the classification tasks with various datasets under both SO(2) and SO(3) rotations. In addition, we further analyze the necessary conditions of applying our approach to PointNet [1] based networks.

High injury severity occurs when a stiff robot arm hits an operator. Introducing compliance into robot systems reduces the impact and enables safe interaction, but at the expense of positioning performance and payload capacity. This paper presents a tunable stiffness mechanism for safe human-robot interaction based on discrete layer jamming. The proposed design of a discrete layer jamming mechanism is a robot link made of multiple thin layers of ABS and multiple clamps. By applying high clamping pressure to the laminates, the link behaves like a rigid link; reducing the clamping pressure softens the link which yields safer human-robot interaction. Compared to conventional pneumatic layer jamming, discrete layer jamming allows for simplicity of installation with dynamic actuators, faster control, greater portability since no air supply is needed, and no sealing issues. To validate the concept, this paper investigates a discrete layer jamming beam made of ten ABS laminates and two aluminum clamps that cover 10% of the surface of the beam. Stiffness tests have been performed, showing that around 17 times bending stiffness change is achieved by increasing the clamping pressure of two clamps from 0 to 1 MPa.

This paper proposes a machine learning method to predict the solutions of related nonlinear optimal control problems given some parametric input, such as the initial state. The map between problem parameters to optimal solutions is called the problem-optimum map, and is often discontinuous due to nonconvexity, discrete homotopy classes, and control switching. This causes difficulties for traditional function approximators such as neural networks, which assume continuity of the underlying function. This paper proposes a mixture of experts (MoE) model composed of a classifier and several regressors, where each regressor is tuned to a particular continuous region. A novel training approach is proposed that trains classifier and regressors independently. MoE greatly outperforms standard neural networks, and achieves highly reliable trajectory prediction (over 99.5% accuracy) in several dynamic vehicle control problems.

We present a novel sound localization algorithm for a non-line-of-sight (NLOS) sound source in indoor environments. Our approach exploits the diffraction properties of sound waves as they bend around a barrier or an obstacle in the scene. We combine a ray tracing-based sound propagation algorithm with a Uniform Theory of Diffraction (UTD) model, which simulate bending effects by placing a virtual sound source on a wedge in the environment. We precompute the wedges of a reconstructed mesh of an indoor scene and use them to generate diffraction acoustic rays to localize the 3D position of the source. Our method identifies the convergence region of those generated acoustic rays as the estimated source position based on a particle filter. We have evaluated our algorithm in multiple scenarios consisting of static and dynamic NLOS sound sources. In our tested cases, our approach can localize a source position with an average accuracy error of 0.7m, measured by the L2 distance between estimated and actual source locations in a 7m×7m×3m room. Furthermore, we observe 37% to 130% improvement in accuracy over a state-of-the-art localization method that does not model diffraction effects, especially when a sound source is not visible to the robot.

Series elastic actuators have proven to be an elegant response to the issue of safety around human-robot interaction. The compliant nature of series elastic actuators provides the potential to be applied in robot-aided rehabilitation for patients with upper and lower limb musculoskeletal injuries. This paper proposes a new series elastic actuator to be used in robot-aided musculoskeletal rehabilitation. The actuator is composed of a DC motor, a torsion spring, and a magnetic particle brake coupled to one common output shaft through a differential gear. The proposed topology focuses on three types of actuation modes most commonly used in rehabilitation, i.e., free motion, elastic, and assistive/resistive motion. A dynamic model of the actuator is presented and validated experimentally and the ability of the actuator to follow a reference torque is shown in different experimental scenarios.

This paper addresses the problem of vision-based robot control where a reference image defines the equilibrium. Specifically, we consider the class of intensity-based nonmetric solutions, which provide for high accuracy, versatility, and robustness. Existing techniques within that class present either a fully coupled control error dynamics or at best only achieve decoupling of the translational part, i.e., they can only obtain a lower triangular system. These couplings in the system dynamics increase analysis complexity and may degrade system performance. This work proposes a new nonlinear observer for also decoupling the rotational part, i.e., for diagonally decoupling the entire control error dynamics. Theoretical proofs of stability and of those decoupling properties are provided. Improved performances are also experimentally confirmed using synthetic and real data, planar and nonplanar objects, simulating and applying a camera-mounted 6-DoF robotic arm.

Dexterous multi-fingered robotic hands can perform a wide range of manipulation skills, making them an appealing component for general-purpose robotic manipulators. However, such hands pose a major challenge for autonomous control, due to the high dimensionality of their configuration space and complex intermittent contact interactions. In this work, we propose deep reinforcement learning (deep RL) as a scalable solution for learning complex, contact rich behaviors with multi-fingered hands. Deep RL provides an end-to-end approach to directly map sensor readings to actions, without the need for task specific models or policy classes. We show that contact-rich manipulation behavior with multi-fingered hands can be learned by directly training with model-free deep RL algorithms in the real world, with minimal additional assumption and without the aid of simulation. We learn to perform a variety of tasks on two different low-cost hardware platforms entirely from scratch, and further study how the learning can be accelerated by using a small number of human demonstrations. Our experiments demonstrate that complex multi-fingered manipulation skills can be learned in the real world in about 4-7 hours for most tasks, and that demonstrations can decrease this to 2-3 hours, indicating that direct deep RL training in the real world is a viable and practical alternative to simulation and model-based control. https:// sites.google.com/view/deeprl-handmanipulation.

The ongoing development of the Inverted PENdulum Type Assistant Robot (I-PENTAR) is being undertaken by the authors. The I-PENTAR has many tasks like lifting objects with unknown mass, pushing and pulling a carts and many more. During execution of these tasks unknown disturbances enter the system and cause dynamic responses and errors. Most notably, the wheels of the robot move with excessively transients. To alleviate this problem, a strain gauge based sensor is attached to the existing structure of the robot to gain information regarding the disturbances. This allows minimal change to the system design while improving robustness to disturbances. In this paper the development of the sensor, the corresponding model and a method of updating the existing purely estimation based control is presented together with tests.

In this paper, a novel, low-cost lower extremity gait rehabilitation device using a single actuator is presented. The proposed device is based on a single DOF 8-bar Jansen mechanism, which was recently introduced as an efficient walking mechanism for legged robots. The mechanism is synthesized to generate the ankle trajectory during human gait relative to the hip, in terms of both position and time. Two mechanisms, one for each lower limb, are applied reciprocally and mechanically synchronized to guarantee symmetric gait. A custom designed seat-type weight support system is also introduced. It supports weight of the user and mechanisms, and also provides the required interaction while maintaining mobility at the hip. To accommodate different users, several parameters of the mechanism are adjustable. Ease of donning-doffing action, weight-bearing and possibility of unhindered arm swing have been considered to provide an effective and user-friendly training environment. A prototype is manufactured, and a pilot study with a healthy subject is conducted to demonstrate feasibility of the concept. Due to ease of control, cost-effectiveness and high intrinsic safety, the proposed system potentially offers a possible method of gait training.

To date, the aim of spinal cord injury (SCI) researches in animals is to find the most effective treatment method which can lead to faster recovery. In order to evaluate if the method is effective, robust functional assessments are crucial. From the past to present, indicators to observe the recovery of the motor function in rodent SCI models are using human observance or the Basso, Beattie, and Bresnahan score (BBB score), force detection, and imaging approaches. Nevertheless, these indicators do not meet some requirements for a severe full transection injury case. The goal of this project is to develop a novel force sensing system for measuring the ground reaction force of rats with severe SCI. In total, this system was tested with 12 spinalized rats. Following a full transection at the T9-T10 level of the spinal cord in rats with a 2mm gap, a nanofiber scaffold containing Neurotrophin-3 (NT-3), as previously described, was implanted [1]. After 12 weeks of rehabilitative training, results showed that rats that underwent rehabilitation were able to gradually exert more force as compared to rats that did not undergo rehabilitation. At Week 6, the ground reaction force recorded in rats with rehabilitation was 0.8 ± 0.1 N in left limb and 0.75 ± 0.14 N in right limb. On the other hand, rats without rehabilitation exerted 0.52 ± 0.06 N in left limb and 0.47 ± 0.09 N in right limb. At Week 12, the force recorded in rehabilitated rats increased to 1.43 ± 0.13 N in left limb and 1.28 ± 0.17 N in right limb whereas in rats without rehabilitation, the force recorded was only 0.74 ± 0.12 N in left limb and 0.54 ± 0.11 N in right limb. These results not only showed that rehabilitation enhanced recovery of motor function, but also demonstrated the viability of measuring the ground reaction force applied by the rats as an assessment for a full spinal cord transection injury model.

Despite the recent advances in soft endoscopes, they could not yet fully fulfill the requirements for minimally invasive and natural orifice transluminal endoscopic surgeries. Maneuverability, bendability, different structural stiffness required for different endoscopic surgical interventions, the space needed for surgical manipulators and patient’s safety are amongst the main factors which can contribute to implementing the new soft robotics endoscope in practice. In this study, based on finite element analysis on an existing endoscopic segment, a new improved endoscopic module was developed. A novel approach for stiffening of the endoscopic module was proposed. The actuation and stiffening components were combined to introduce a multi-level stiffening mechanism to the endoscope, and also to provide a free lumen for manipulators. To increase patient’s safety, a force sensing module was developed to estimate the magnitude and direction of the force from tissues to the endoscope. The developed endoscopic system was integrated to a haptic control system. The 3D kinematics control and haptic feedback control of the endoscopic module were validated.

The purpose of this work is to introduce a newly developed exoskeleton robot designed to characterize the neuromuscular properties of the shoulder, including intrinsic and reflexive mechanisms, during static posture and dynamic movement in a 3-dimensional space. Quantitative characterization of these properties requires fast perturbation (>100°/s) to separate their contribution from that of voluntary mechanism. Understanding these properties of the shoulder control could assist in the rehabilitation or enhancement of upper limb performance during physical human-robot interaction. The device can be described as a new type of spherical parallel manipulator (SPM) that utilizes three 4-bar (4B) substructures to decouple and control roll, pitch and yaw of the shoulder. By utilizing a parallel architecture, the 4BSPM exoskeleton has the advantage of high acceleration, fast enough to satisfy the speed requirement for the characterization of distinct neuromuscular properties of the shoulder. In this work, the prototype is presented, along with an evaluation of its position accuracy and step response. The development and preliminary testing of the 4B-SPM exoskeleton presented in this work demonstrates its potential to be a useful tool for studying the neuromuscular mechanisms of the shoulder joint.

High risk of a collision between rotor blades and the obstacles in a complex environment imposes restrictions on the aerial manipulators. To solve this issue, a novel system cable-Suspended Aerial Manipulator (SAM) is presented in this paper. Instead of attaching a robotic manipulator directly to an aerial carrier, it is mounted on an active platform which is suspended on the carrier by means of a cable. As a result, higher safety can be achieved because the aerial carrier can keep a distance from the obstacles. For self-stabilization, the SAM is equipped with two actuation systems: winches and propulsion units. This paper presents an overview of the SAM including the concept behind, hardware realization, control strategy, and the first experimental results.

Traditional environmental and structural monitoring often uses static sensor networks deployed at predetermined locations or mobile robots that use a rastering technique for area coverage. These methods rely on the operators making assumptions about the nature of the unknown field that is being measured and are often time-consuming for localizing an area of interest. Here, we aim to quickly localize possible leaks within high-level nuclear waste tanks at the Hanford facility. The structure of these tanks precludes most sensor network approaches and raises many issues with robotic inspection, such as navigation within highly constrained environments. This work uses a Bayesian Optimization approach for guiding a mobile robot's search strategy and implements a utility function that allows for prior knowledge of the structure to be incorporated when selecting future search locations. Compared to traditional exhaustive approaches, our method quickly reduces RMSE error and shortens the distance the robot must travel.

Mechanical stresses on the spine are a significant risk factor for low back pain, a highly prevalent health problem around the world. Certain occupational activities such as repetitive heavy lifting and static bending posture lead to high loads on the lower back. To address this problem, we are developing a soft power suit capable of reducing physical load on the lower back during dynamic lifting and static forward bending. The power suit is designed to mimic the force transmission in the body and duplicate the force generated by muscles and tendons. Two twisted string actuators (TSAs) attached to a back brace are used to generate tensile forces which assist the underlying muscles to control trunk flexion. The fabric construction and TSA enable a lightweight design of the suit: without the battery, the entire system weighs only 2.4kg. Here we present the design and implementation of the prototype system along with a preliminary biomechanical study that evaluates the effect of the system on the body. The results show that the power suit does not change the wearer’s bending kinematics and helps the subject to keep the static bending posture. Moreover, using the power suit significantly reduced the muscle activation required for both static bending and dynamic lifting (50.2–54.0% and 21.4–25.2% reduction, respectively).

Recent researches demonstrate that selflocalization performance is a very useful measure of likelihood-of-change (LoC) for change detection. In this paper, this “detection-by-localization” scheme is studied in a novel generalized task of object-level change detection. In our framework, a given query image is segmented into object-level subimages (termed “scene parts”), which are then converted to subimagelevel pixel-wise LoC maps via the detection-by-localization scheme. Our approach models a self-localization system as a ranking function, outputting a ranked list of reference images, without requiring relevance score. Thanks to this new setting, we can generalize our approach to a broad class of selflocalization systems. We further propose an aggregation of different self-localization results from different queries so as to achieve higher precision. Our ranking based self-localization model allows to fuse self-localization results from different modalities via an unsupervised rank fusion derived from a field of multi-modal information retrieval (MMR). Our framework does not rely on the raw-score-merging hypothesis. Challenging experiments of cross-season change detection using the publicly available North Campus Long-Term (NCLT) dataset validates the efficacy of our proposed method.