ICRA 2019 论文列表

Neuromorphic computing is a new paradigm for design of both the computing hardware and algorithms inspired by biological neural networks. The event-based nature and the inherent parallelism make neuromorphic computing a promising paradigm for building efficient neural network based architectures for control of fast and agile robots. In this paper, we present a spiking neural network architecture that uses sensory feedback to control rotational velocity of a robotic vehicle. When the velocity reaches the target value, the mapping from the target velocity of the vehicle to the correct motor command, both represented in the spiking neural network on the neuromorphic device, is autonomously stored on the device using on-chip plastic synaptic weights. We validate the controller using a wheel motor of a miniature mobile vehicle and inertia measurement unit as the sensory feedback and demonstrate online learning of a simple “inverse model” in a two-layer spiking neural network on the neuromorphic chip. The prototype neuromorphic device that features 256 spiking neurons allows us to realise a simple proof of concept architecture for the purely neuromorphic motor control and learning. The architecture can be easily scaled-up if a larger neuromorphic device is available.

With the proliferation of small aerial vehicles, acquiring close up imagery for high quality reconstruction is gaining importance. We present an adaptive view planning method to collect such images in an automated fashion. We first start by sampling a small set of views to build a coarse proxy to the scene. We then present (i) a method that builds a set of adaptive viewing planes for efficient view selection and (ii) an algorithm to plan a trajectory that guarantees high reconstruction quality which does not deviate too much from the optimal one. The vehicle then follows the trajectory to cover the scene, and the procedure is repeated until reconstruction quality converges or a desired level of quality is achieved. The set of viewing planes provides an effective compromise between using the entire 3D free space and using a single view hemisphere to select the views. We compare our algorithm to existing methods in three challenging scenes. Our algorithm generates views which produce the least reconstruction error comparing to three different baseline approaches.

Robots that are trained to perform a task in a fixed environment often fail when facing unexpected changes to the environment due to a lack of exploration. We propose a principled way to adapt the policy for better exploration in changing sparse-reward environments. Unlike previous works which explicitly model environmental changes, we analyze the relationship between the value function and the optimal exploration for a Gaussian-parameterized policy and show that our theory leads to an effective strategy for adjusting the variance of the policy, enabling fast adapt to changes in a variety of sparse-reward environments.

Probabilistic vehicle trajectory prediction is essential for robust safety of autonomous driving. Current methods for long-term trajectory prediction cannot guarantee the physical feasibility of predicted distribution. Moreover, their models cannot adapt to the driving policy of the predicted target human driver. In this work, we propose to overcome these two shortcomings by a Bayesian recurrent neural network model consisting of Bayesian-neural-network-based policy model and known physical model of the scenario. Bayesian neural network can ensemble complicated output distribution, enabling rich family of trajectory distribution. The embedded physical model ensures feasibility of the distribution. Moreover, the adopted gradient-based training method allows direct optimization for better performance in long prediction horizon. Furthermore, a particle-filter-based parameter adaptation algorithm is designed to adapt the policy Bayesian neural network to the predicted target online. Effectiveness of the proposed methods is verified with a toy example with multi-modal stochastic feedback gain and naturalistic car following data.

A space robot working in a controlled-floating mode can be used for performing in-orbit telescope assembly through simultaneously controlling the motion of the spacecraft base and its robotic arm. Handling and assembling optical mirrors requires the space robot to achieve slow and precise manoeuvres regardless of the disturbances and errors in the trajectory. The robustness offered by the nonlinear H∞ controller, in the presence of environmental disturbances and parametric uncertainties, makes it a viable solution. However, using fixed tuning parameters for this controller does not always result in the desired performance as the arm's trajectory is not known a priori for orbital assembly missions. In this paper, a complete study on the impact of the different tuning parameters is performed and a new adaptive H∞ controller is developed based on bounded functions. The simulation results presented show that the proposed adaptive H∞ controller guarantees robustness and precise tracking using a minimal amount of forces and torques for assembly operations using a small space robot.

Neuroevolution is a process of training neural networks (NN) through an evolutionary algorithm, usually to serve as a state-to-action mapping model in control or reinforcement learning-type problems. This paper builds on the Neuro Evolution of Augmented Topologies (NEAT) formalism that allows designing topology and weight evolving NNs. Fundamental advancements are made to the neuroevolution process to address premature stagnation and convergence issues, central among which is the incorporation of automated mechanisms to control the population diversity and average fitness improvement within the neuroevolution process. Insights into the performance and efficiency of the new algorithm is obtained by evaluating it on three benchmark problems from the Open AI platform and an Unmanned Aerial Vehicle (UAV) collision avoidance problem.

Lower-limb exoskeleton has gained considerable interests in walking assistance applications for paraplegic patients. In walking assistance of paraplegic patients, the exoskeleton should have the ability to help patients to walk over different terrains in the daily life, such as slope terrains. One critical issue is how to plan the stepping locations on slopes with different gradients, and generate stable and human-like gaits for patients. This paper proposed an adaptive gait planning approach which can generate gait trajectories adapt to slopes with different gradients for lower-limb walking assistance exoskeletons. We modeled the human-exoskeleton system as a 2D Linear Inverted Pendulum Model (2D-LIPM) with an external force in the two-dimensional sagittal plane, and proposed a Dynamic Gait Generator (DGG) based on an extension of the conventional Capture Point (CP) theory and Dynamic Movement Primitives (DMPs). The proposed approach can dynamically generate reference foot locations for each step on slopes, and human-like adaptive gait trajectories can be reproduced after the learning from demonstrated trajectories that sampled from level ground walking of normal healthy human. We demonstrated the efficiency of the proposed approach on both the Gazebo simulation platform and an exoskeleton named AIDER. Experimental results indicate that the proposed approach is able to provide the ability for exoskeletons to generate appropriate gaits adapt to slopes with different gradients.

This paper is concerned with the optimal kinematic control of a robot manipulator where the robot end effector position follows a task space trajectory. The joints are actuated with the desired velocity profile to achieve this task. This problem has been solved using a single network adaptive critic (SNAC) by expressing the forward kinematics as input affine system. Usually in SNAC, the critic weights are updated using back propagation algorithm while little attention is given to convergence to the optimal cost. In this paper, we propose a critic weight update law that ensures convergence to the desired optimal cost while guaranteeing the stability of the closed loop kinematic control. In kinematic control, the robot is required to reach a specific target position. This has been solved as an optimal regulation problem in the context of SNAC based kinematic control. When the robot is required to follow a time varying task space trajectory, then the kinematic control has been framed as an optimal tracking problem. For tracking, an augmented system consisting of tracking error and reference trajectory is constructed and the optimal control policy is derived using SNAC framework. The stability and performance of the system under the proposed novel weight tuning law is guaranteed using Lyapunov approach. The proposed kinematic control scheme has been validated in simulations and experimentally executed using a real six degrees of freedom (DOF) Universal Robot (UR) 10 manipulator.

One of the significant challenges of moving from manual to robot-assisted retinal surgery is the loss of perception of forces applied to the sclera (sclera forces) by the surgical tools. This damping of force feedback is primarily due to the stiffness and inertia of the robot. The diminished perception of tool-to-eye interactions might put the eye tissue at high risk of injury due to excessive sclera forces or extreme insertion of the tool into the eye. In the present study therefore a 1-dimensional adaptive control method is customized for 3-dimensional control of sclera force components and tool insertion depth and then implemented on the velocity-controlled Johns Hopkins Steady-Hand Eye Robot. The control method enables the robot to perform autonomous motions to make the sclera force and/or insertion depth of the tool tip to follow pre-defined desired and safe trajectories when they exceed safe bounds. A robotic light pipe holding application in retinal surgery is also investigated using the adaptive control method. The implementation results indicate that the adaptive control is able to achieve the imposed safety margins and prevent sclera forces and insertion depth from exceeding safe boundaries.

We present a study of the dynamics and control of a 28-gram quadrotor during the execution of aerobatic maneuvers in the presence of propeller-aerodynamic-coefficient and torque-latency time-variations. First, through a momentum-theory-based analysis of the flow field surrounding the robot during aerobatic flight, we develop a dynamic linear time-varying (LTV) description of the torque acting on the flyer in which both considered effects explicitly appear as distinct mathematical terms. Then, an adaptive control scheme, composed of a backstepping controller and a modified recursive least-squares (RLS) estimator, is designed to counteract the negative effects produced by the time-varying dynamics of the torque that drives the flyer. The suitability and efficacy of the proposed methods are demonstrated through real-time flight experiments in which the quadrotor autonomously performs three different types of aerobatic maneuvers: triple flips, Pugachev's Cobras and mixed flips. Furthermore, analyses of the experimental data compellingly show that the proposed control scheme consistently improves the performance of the aerial vehicle during aerobatic flight, compared to those achieved by using a high-performance linear time-invariant (LTI) controller that does not account for time-varying torque generation.

Filter-based methods are a suitable option to deal with the burdensome 3D pose estimation problems for their incremental properties. The classical approaches use the Gaussian distribution to model the uncertainty of the pose parameters and recent work has begun to take advantage of the Bingham distribution, which is theoretically more suitable for modeling uncertainty on the SO(3) group. However, these algorithms are still at the very beginning and heavily rely on manual tuning. In this work we equip the Bingham distribution based filter with adaptive filtering ability, which realizes completely autonomous tuning without human interaction. The experiments demonstrate that our method is significantly laborsaving compared to the state-of-the-art methods as well as capable of maintaining high accuracy.

We address the problem of executing tool-using manipulation skills in scenarios where the objects to be used may vary. We assume that point clouds of the tool and target object can be obtained, but no interpretation or further knowledge about these objects is provided. The system must interpret the point clouds and decide how to use the tool to complete a manipulation task with a target object; this means it must adjust motion trajectories appropriately to complete the task. We tackle three everyday manipulations: scraping material from a tool into a container, cutting, and scooping from a container. Our solution encodes these manipulation skills in a generic way, with parameters that can be filled in at run-time via queries to a robot perception module; the perception module abstracts the functional parts of the tool and extracts key parameters that are needed for the task. The approach is evaluated in simulation and with selected examples on a PR2 robot.

In safety-critical environments, robots need to reliably recognize human activity to be effective and trust-worthy partners. Since most human activity recognition (HAR) approaches rely on unimodal sensor data (e.g. motion capture or wearable sensors), it is unclear how the relationship between the sensor modality and motion granularity (e.g. gross or fine) of the activities impacts classification accuracy. To our knowledge, we are the first to investigate the efficacy of using motion capture as compared to wearable sensor data for recognizing human motion in manufacturing settings. We introduce the UCSD-MIT Human Motion dataset, composed of two assembly tasks that entail either gross or fine-grained motion. For both tasks, we compared the accuracy of a Vicon motion capture system to a Myo armband using three widely used HAR algorithms. We found that motion capture yielded higher accuracy than the wearable sensor for gross motion recognition (up to 36.95%), while the wearable sensor yielded higher accuracy for fine-grained motion (up to 28.06%). These results suggest that these sensor modalities are complementary, and that robots may benefit from systems that utilize multiple modalities to simultaneously, but independently, detect gross and fine-grained motion. Our findings will help guide researchers in numerous fields of robotics including learning from demonstration and grasping to effectively choose sensor modalities that are most suitable for their applications.

Robot navigation in outdoor environments is exposed to detrimental factors such as vibrations or power consumption due to the different terrains on which the robot navigates. In this paper, we address the problem of actively improving navigation by planning paths that aim at reducing over time phenomena such as vibrations during traversal. Our approach uses a Gaussian Process (GP) mixture model and an aerial image of the environment to learn and improve continuously a place-dependent model of such phenomena from the experiences of the robot. We use this model to plan paths that trade-off the exploration of unknown promising regions and the exploitation of known areas where the impact of the detrimental factors on navigation is low, leading to an improved navigation over time. We implemented our approach and thoroughly tested it using real-world data. Our experiments suggest that our approach with no initial information leads the robot, after few runs, to follow paths along which it experiences similar vibrations or energy consumption as if it was following the optimal path computed given the ground truth information.

Learning the dynamical system (DS) model from data that preserves dynamical system properties is an important problem in many robot learning applications. Typically, the joint data coming from cyber-physical systems, such as robots have some underlying DS properties associated with it, e.g., convergence, invariance to a set, etc. In this paper, a model learning method is developed such that the trajectories of the DS are invariant in a given compact set. Such invariant DS models can be used to generate trajectories of the robot that will always remain in a prescribed set. In order to achieve invariance to a set, Barrier certificates are employed. The DS is approximated using Extreme Learning Machine (ELM), and a parameter learning problem subject to Barrier certificates enforced at all the points in the prescribed set is solved. To solve an infinite constraint problem for enforcing Barrier Certificates at every point in a given compact set, a modified constraint is developed that is sufficient to hold the Barrier certificates in the entire set. An active sampling strategy is formulated to minimize the number of constraints in learning. Simulation results of ELM learning with and without Barrier certificates are presented which show the invariance property being preserved in the ELM learning when learning procedure involves Barrier constraints. The method is validated using experiments conducted on a robot arm recreating invariant trajectories inside a prescribed set.

We pose an active perception problem where an autonomous agent actively interacts with a second agent with potentially adversarial behaviors. Given the uncertainty in the intent of the other agent, the objective is to collect further evidence to help discriminate potential threats. The main technical challenges are the partial observability of the agent intent, the adversary modeling, and the corresponding uncertainty modeling. Note that an adversary agent may act to mislead the autonomous agent by using a deceptive strategy that is learned from past experiences. We propose an approach that combines belief space planning, generative adversary modeling, and maximum entropy reinforcement learning to obtain a stochastic belief space policy. By accounting for various adversarial behaviors in the simulation framework and minimizing the predictability of the autonomous agent's action, the resulting policy is more robust to unmodeled adversarial strategies. This improved robustness is empirically shown against an adversary that adapts to and exploits the autonomous agent's policy when compared with a standard Chance-Constraint Partially Observable Markov Decision Process robust approach.

In the present work, we propose an active tactile exploration framework to obtain a surface model of an unknown object utilizing multiple contacts simultaneously. To incorporate these multiple contacts, the exploration strategy is based on the differential entropy of the underlying Gaussian process implicit surface model, which formalizes the exploration with multiple contacts within an information theoretic context and additionally allows for nonmyopic multi-step planning. In contrast to many previous approaches, the robot continuously slides along the surface with its end-effectors to gather the tactile stimuli, instead of touching it at discrete locations. This is realized by closely integrating the surface model into the compliant controller framework. Furthermore, we extend our recently proposed sliding based tactile exploration approach to handle non-convex objects. In the experiments, it is shown that multiple contacts simultaneously leads to a more efficient exploration of complex, non-convex objects, not only in terms of time, but also with respect to the total moved distance of all end-effectors. Finally, we demonstrate our methodology with a real PR2 robot that explores an object with both of its arms.

This work is a preliminary study assessing the feasibility of using cold-gas thrusters for active damping of flexible cable-driven parallel robots. The concept is validated experimentally on a planar robot embedding custom-built supersonic air thrusters operating at an industry-standard pressure level.

Recent advances in teleoperation-based robotic-assisted Minimally Invasive Surgery (MIS) have made significant inroads in clinical adoption. However, such master-slave surgical systems create a physical separation between the surgeon and the patient. The concept of Active Constraints (ACs) provides guidance and sensory information to surgical robot operators in a form of haptic, visual or audible cues. This work proposes a novel ACs approach to avoid surgical tool-clashing and collision of the tool-shaft with delicate anatomy using elasto-plastic frictional force control. The presented framework is designed to reduce the occurence of direct coupling during electrocautery and to protect high-risk regions in Minimally Invasive Partial Nephrectomy (MIPN). Moreover, we combine aforementioned ACs methods and propose a solution when simultaneous penetration of both constraints occurs. The proposed methodology is implemented on the teleoperated da Vinci Surgical System using the da Vinci Research Kit (dVRK) and its performance is demonstrated through three types of user experiments. The experimental results show that the developed algorithms are of significant benefit in performing the tasks with ACs assistance.