ICRA 2019 Papers

Supernumerary Robotic Limbs (SRLs) are wearable robots augmenting human capabilities by acting as a co-worker, reaching objects, support human arms, etc. However, existing SRLs lack the mechanical backdrivability and bandwidth required for tasks where the interaction forces must be controlled such as painting, drilling, manipulating fragile objects, etc. Being highly backdrivable with a high bandwidth while minimizing weight presents a major technological challenge imposed by the limited performances of conventional electromagnetic actuators. This paper studies the feasibility of using magnetorheological (MR) clutches coupled-to a low-friction hydrostatic transmission to provide a highly capable, but yet lightweight, force-controllable SRL. A 2.7 kg 2-DOFs wearable robotic arm is designed and built. Shoulder and elbow joints are designed to deliver 39 and 25 Nm, with 115 and 180° of range of motion. Experimental studies conducted on a one-DOF test bench and validated analytically demonstrate a high force bandwidth (>25 Hz) and a good ability to control interaction forces even when interacting with an external impedance. Furthermore, three force-control approaches are studied and demonstrated experimentally: open-loop, closed-loop on force, and closed-loop on pressure. All three methods are shown to be effective. Overall, the proposed MR-Hydrostatic actuation system is well-suited for a lightweight SRL interacting with both human and environment that add unpredictable disturbances.

Automatic assembly has broad applications in industries. Traditional assembly tasks utilize predefined trajectories or tuned force control parameters, which make the automatic assembly time-consuming, difficult to generalize, and not robust to uncertainties. In this paper, we propose a learning framework for high precision industrial assembly. The framework combines both the supervised learning and the reinforcement learning. The supervised learning utilizes trajectory optimization to provide the initial guidance to the policy, while the reinforcement learning utilizes actor-critic algorithm to establish the evaluation system even the supervisor is not accurate. The proposed learning framework is more efficient compared with the reinforcement learning and achieves better stability performance than the supervised learning. The effectiveness of the method is verified by both the simulation and experiment. Experimental videos are available at [1].

This paper presents the development of a fiber Bragg grating (FBG) bending sensor for shape memory alloy (SMA) bending modules. Due to the small form factor, low cost, and large-deflection capability, SMA bending modules can be used to construct disposable surgical robots for a variety of minimally invasive procedures. To realize a closed-loop control of SMA bending modules, an intrinsic bending sensor is imperative. Due to the lack of bending sensors for SMA bending modules, we have developed an FBG bending sensor by integrating FBG fibers with a superelastic substrate using flexible adhesive. Since the substrate is ultra-thin and adhesive is flexible, the sensor has low stiffness and can measure large curvatures. Additionally, due to the orthogonal arrangement of the sensor/actuator assembly, the influence of temperature variation caused by SMA actuation can be compensated. The working principle of the developed sensor was modeled followed by simulations. After experimentally evaluating the developed model, the sensor was integrated with an SMA bending module and cyclically bi-directionally deflected. The experimental results proved the relatively high measurement accuracy, high repeatability, and large measurable curvatures of the sensor, although hysteresis was observed due to friction.

The ability to accurately measure signal time-of-flight between ultra-wideband (UWB) wireless communication transceivers, even in multipath environments, makes this technology ideally suited to develop ranging-based positioning systems, especially for indoor applications where GPS signals are not available. In recent years, low-cost commercial UWB transceivers have become more easily available and increasingly used to develop custom robot positioning systems. In this paper, we focus in particular on positioning techniques requiring the synchronization of base stations such as Time of Arrival (TOA) and Time Difference of Arrival (TDOA). We present a protocol based on Kalman filtering for simultaneous synchronization of multiple UWB base stations and positioning of an arbitrary number of passive UWB receivers. We illustrate experimentally using our protocol and an EKF-based navigation system design the level of accuracy achievable with small low-power UWB modules for mobile robot positioning. We discuss in details measurement errors and system tuning issues applicable to popular commercial UWB transceivers.

In this paper, we study the cooperative path planning and motion coordination problems of the multi-robot system with large number of robots, aiming for practical applications in robotic warehouses and automated transportation systems. Particularly, we solve the life-long planning problem and guarantee the coordination performance in the presence of robot motion uncertainties. A hierarchical path planning and motion coordination structure is presented. The environment is divided into several sectors and a traffic heat-map is presented to describe the current sector-level traffic condition. In path planning level, the sector-level path is calculated by considering the path distance, the current traffic condition and the current robot uncertainty. In motion coordination level, local cooperative A* algorithm and conflict-based searching strategy are utilized within each sector to generate the collision-free local path of each robot in a rolling planning manner. The effectiveness and practical applicability of the proposed approach are validated by simulations with more than one thousand robots and real experiments.

The article proposes a new heuristic for task allocation and routing of heterogeneous robots. Specifically, we consider a path planning problem where there are two (structurally) heterogeneous robots that start from distinctive depots and a set of targets to visit. The objective is to find a tour for each robot in a manner that enables each target location to be visited at least once by one of the robots while minimizing the maximum travel cost. A solution for Multiple Depot Heterogeneous Traveling Salesman Problem (MDHTSP) with min-max objective is in great demand with many potential applications, because it can significantly reduce the job completion duration. However, there are still no reliable algorithms that can run in short amount of time. As an initial idea of solving min-max MDHTSP, we present a heuristic based on a primal-dual technique that solves for a case involving two robots while focusing on task allocation. Based on computational results of the implementation, we show that the proposed algorithm produces a good quality of feasible solution within a relatively short computation time.

Typically, the problem of robotic manipulation is divided among two sequential phases: a planning one and an execution one. However, since the second one is executed in open loop, the robot is unable to react in real time to changes in the task (e.g. moving object). This paper addresses the mobile manipulation problem from a real-time, closed loop perspective. In particular, we propose a generic optimization-based Cartesian controller, that given a continuous monitoring of the goal, determines the best motion commands. We target our controller to a robotic system comprising an arm and a mobile platform. However, the approach can in principle be extended to more complex mechanisms. The approach is based on shifting the problem to velocity space, where end effector velocity is a linear function of joint and base platform velocities. Our approach was quantitatively evaluated both on simulation and on a real service robot. It was also integrated into a mobile service robot architecture targeting domestic tasks and evaluated on the RoboCup@Home scientific competition. Our results show that the controller is able to reach random arm configurations with a high probability of success.

The choice of model orientation is a very important issue in Additive Manufacturing (AM). In this paper, the model orientation problem is formulated as a multi-objective optimization problem, aiming at minimizing the building time, the surface quality, and the supporting area. Then we convert the problem into a single-objective optimization in the linear-weighted way. After that, the Genetic Algorithm (GA) is used to solve the optimization problem and the process of GA is parallelized and implemented on GPU. Experimental results show that when dealing with complex models in AM, compared with CPU only implementation, the GPU based GA can speed up the process by about 50 times, which helps to significantly reduce the optimization time and ensure the quality of solutions. The GPU based parallel methods we proposed can help to reduce the execution time and improve the efficiency greatly, making the processes more efficient.

Robots that perform autonomous maneuvering in a disaster environment usually dont have perfect understanding of the environment in advance. The robot is continuously evaluating the environment as it proceeds, deciding the optimal way to traverse the environment to get to the goal. A critical aspect of this decision is the robot estimate of the terrain accessibility index, which quantifies how easy it is to navigate through the immediate terrain. This paper represents a new method to calculate terrain accessibility index based on obstacle distance to the robots position. In addition, a Fuzzy Inference System Vector Field Histogram (FISVFH) method has been designed for automating the selection of the robots linear and angular velocities in the VFH (Vector Field Histogram) algorithm, based on the calculated sector accessibility index. The proposed method is a two-input and two-output Fuzzy Inference System, where the current robot heading, and sector accessibility index serve as the input, and the corresponding linear and angular velocities to the VFH algorithm are outputs. The VFH, VFH + and FISVFH are tested both in simulation and experimentation in 4 environments that are known to result in failures in VFH and VFH +, for comparison purposes. In addition, the algorithm was verified through experimental setup of a disaster prone environment. In both simulation and experimentation the results show that FISVFH outperforms VFH and VFH +. It is also shown that the FISVFH algorithm is capable of handling the disaster prone environment. Overall, the FISVFH algorithm enables the robot to get to the goal faster and also produces a smoother path while doing so.

Skid-steer drive systems are widely used in mobile robot platforms. Such systems are subject to significant slippage and skidding during normal operation due to their nature. The ability to predict and compensate for such slippages in the forward kinematics of these types of robots is of great importance and provides the means for accurate control and safe navigation. In this work, we propose a new kinematic model capable of slip prediction for skid-steer wheeled mobile robots (SSWMRs). The proposed model outperforms the state-of-the-art in terms of both translational and rotational prediction error on a dataset composed of more than 6 km worth of trajectories traversed by a skid-steer robot. We also publicly release our dataset to serve as a benchmark for system identification and model learning research for SSWMRs.

The development of high-performance perception for mobile robotic agents is still challenging. Learning appropriate perception models usually requires extensive amounts of labeled training data that ideally follows the same distribution as the data an agent will encounter in its target task. Recent developments in gaming industry led to game engines able to generate photorealistic environments in real-time, which can be used to realistically simulate the sensory input of an agent.We propose a novel framework which allows the definition of different learning scenarios and instantiates these scenarios in a high quality game engine where a perceptual agent can act and learn in. The scenarios are specified in a newly developed scenario description language that allows the parametrization of the virtual environment and the perceptual agent. New scenarios can be sampled from a task-specific object distribution that allows the automatic generation of extensive amounts of different learning environments for the perceptual agent.We will demonstrate the plausibility of the framework by conducting object recognition experiments on a real robotic system which has been trained within our framework.

In this paper we introduce a novel framework for expressing and learning force-sensitive robot manipulation skills. It is based on a formalism that extends our previous work on adaptive impedance control with meta parameter learning and compatible skill specifications. This way the system is also able to make use of abstract expert knowledge by incorporating process descriptions and quality evaluation metrics. We evaluate various state-of-the-art schemes for meta parameter learning and experimentally compare selected ones. Our results clearly indicate that the combination of our adaptive impedance controller with a carefully defined skill formalism significantly reduces the complexity of manipulation tasks even for learning peg-in-hole with submillimeter industrial tolerances. Overall, the considered system is able to learn variations of this skill in under 20 minutes. In fact, experimentally the system was able to perform the learned tasks without visual feedback faster than humans, leading to the first learning-based solution of complex assembly at such real-world performance.

Learning the dynamics of robots from data can help achieve more accurate tracking controllers, or aid their navigation algorithms. However, when the actual dynamics of the robots change due to external conditions, on-line adaptation of their models is required to maintain high fidelity performance. In this work, a framework for on-line learning of robot dynamics is developed to adapt to such changes. The proposed framework employs an incremental support vector regression method to learn the model sequentially from data streams. In combination with the incremental learning, strategies for including and forgetting data are developed to obtain better generalization over the whole state space. The framework is tested in simulation and real experimental scenarios demonstrating its adaptation capabilities to changes in the robot's dynamics.

This paper presents a novel permanent magnets based actuator capable of injecting and 2D wireless control the motion of an untethered microrobot. The actuator is obtained with a simple, but an effective arrangement of four permanent magnets. Its novelty is that it creates local maxima of the magnetic field magnitude in a planar workspace. This results in a convergence point for magnetic microrobots that are in its influence zone. Trapping microrobots in the local maxima makes their open-loop guidance possible, even in presence of reasonable perturbations. Actually, open-loop control is the only way to achieve some drug delivery when there are no sensors to provide us a feedback on the microrobot's positions. With the proposed actuator, the workspace is reduced, the movement is simplified and the actuator orientation is not needed anymore. Experimental results are provided in the paper to show the soundness of the proposed actuator design.

The growing demand of industrial, automotive and service robots presents a challenge to the centralized Cloud Robotics model in terms of privacy, security, latency, bandwidth, and reliability. In this paper, we present a `Fog Robotics' approach to deep robot learning that distributes compute, storage and networking resources between the Cloud and the Edge in a federated manner. Deep models are trained on non-private (public) synthetic images in the Cloud; the models are adapted to the private real images of the environment at the Edge within a trusted network and subsequently, deployed as a service for low-latency and secure inference/prediction for other robots in the network. We apply this approach to surface decluttering, where a mobile robot picks and sorts objects from a cluttered floor by learning a deep object recognition and a grasp planning model. Experiments suggest that Fog Robotics can improve performance by sim-to-real domain adaptation in comparison to exclusively using Cloud or Edge resources, while reducing the inference cycle time by 4× to successfully declutter 86% of objects over 213 attempts.

Cloud Robotics is a paradigm where multiple robots are connected to cloud services via Internet to access “unlimited” computation power, at the cost of network communication. However, due to limitations such as network latency and variability, it is difficult to control dynamic, human compliant service robots directly from the cloud. In this work, we combine cloud robotics with an agile edge device to build a Fog Robotic system by leveraging an asynchronous protocol with a “heartbeat” signal. We use the system to enable robust teleoperation of a dynamic self-balancing robot from the cloud. We use the system to pick up boxes from static locations, a task commonly performed in warehouse logistics. To make cloud teleoperation more intuitive and efficient, we program a cloud-based image based visual servoing (IBVS) module to automatically assist the cloud teleoperator during the object pickups. Visual feedbacks, including apriltag recognition and tracking, are performed in the cloud to emulate a Fog Robotic object recognition system for IBVS. We demonstrate the feasibility of a dynamic real-time automation system using this cloud-edge hybrid design, which opens up possibilities of deploying dynamic robotic control with deep-learning recognition systems in Fog Robotics. Finally, we show that Fog Robotics enables the self-balancing service robot to pick up a box automatically from a person under unstructured environments.

Dexterous, serial-chain motor-driven robotic arms have high moving mass, since most of the actuators must be located in the arm itself. This necessitates high gear ratios, sacrificing passive compliance, backdrivability, and the capacity for delicate motion. We introduce the concept of a remote direct-drive (RDD) manipulator, in which every motor is located in the base, connected to remote joints via a low-friction hydrostatic transmission. We have designed a new hydrostatic linear actuator with a fully-floating piston; the piston floats within the cylinder using a pair of soft fiber-elastomer rolling-diaphragm seals. This eliminates static friction from seal rubbing and piston/rod misalignment. Actuators were developed with a 20mm bore, weighing 55 grams each with a 400:1 bidirectional strength-to-weight ratio ( + /-230\mathrm {N}( + /-230\mathrm {N}), which drive a 2-DOF manipulator (wrist pitch/finger pinch; 120-degree range-of-motion; 6.6 Nm max grip strength). The gripper is hydrostatically coupled to remotely-located direct-drive/backdrivable brushless electric motors. System hysteresis and friction are 1 percent of full-range force. This low-mass low-friction configuration is of great interest for powered prosthetic hand design, and passively-safe high dynamic range robot arms.

In this paper we present a flexible low-cost sonar sensor platform that can be used for a wide range of biomimetic sonar experiments and autonomous sonar navigation targeted at robotics applications. The navigation abilities of bats using ultrasound (sonar) in unknown cluttered environments are very effective and can be distilled into a sensor architecture and accompanying control methodology that lends itself to be implemented on cost efficient hardware. The sensor architecture and processing methodology of this sensing platform mimics that of bats. In this paper we specifically focused on the common big-eared bat (Micronycteris microtis) although this could be transferred to other bat species or even other echolocating animals since the experimental platform was designed for flexibility. Using this platform we were able to implement a control system using a subsumption architecture that features different behavior patterns based solely on the sonar sensor as a source of exteroceptive information. In order to validate the combination of our autonomous navigation control system and our developed sonar sensor platform, the hardware was mounted on the P3DX robotics platform that was introduced in an unknown testing environment and have it drive autonomously. These experiments were used to validate our assumption of the efficacy of these relatively simple biomimetic control mechanisms and thus alleviating the need for expensive sensing platforms for certain robotics applications.

We introduce a unique experimental testbed that consists of a fleet of 16 miniature Ackermann-steering vehicles. We are motivated by a lack of available low-cost platforms to support research and education in multi-car navigation and trajectory planning. This article elaborates the design of our miniature robotic car, the Cambridge Minicar, as well as the fleet's control architecture. Our experimental testbed allows us to implement state-of-the-art driver models as well as autonomous control strategies, and test their validity in a real, physical multi-lane setup. Through experiments on our miniature highway, we are able to tangibly demonstrate the benefits of cooperative driving on multi-lane road topographies. Our setup paves the way for indoor large-fleet experimental research.