ICRA 2017 论文列表

Magnetic interaction can be utilized for remote actuation of surgical manipulators. However, platforms currently available for that purpose have limited workspaces, inadequate field strength or very low bandwidth of the electrical subsystem. In this paper, we present BigMag, a novel platform capable of magnetic steering of continuum manipulators for medical purposes. BigMag comprises of 6 mobile coils and is capable of generating the fields of at least 40 mT in any direction at every point of its workspace. Moreover, we introduce a mathematical model for 3D mobile coil arrays. Each coil is modelled using finite element data adjusted by measurement-based correction, (a maximum observed mean error between the model and the prediction of 3.36 ± 5.62%). The model for a full system is validated in two tasks. In the first task, the system executes a prescribed rotating field (mean error between the model and measurement of 7.51% and minimum R2 of 0.964). The second task tests the estimation of the field for known 3D trajectories (minimum R2 of 0.967). The investigation concludes with a demonstration of BigMag capabilities in actuation of magnetic catheters in confined spaces usinguser-controlled steering.

There has been consistent trials to make a hydraulic actuator to be effective in interactive robot applications that require high power and durability. However, conventional strategies to obtain this goal, such as force/torque servo control or adding elastic components, suffer from complexity, sensor problems, and reduced system robustness. In this paper, we show that introducing backdrivability to hydraulic system may be an ultimate solution to constructing interactive hydraulic robot systems, by analyzing the advantages of backdrivability for this application. Moreover, we show that the backdrivability of valve-controlled hydraulic system can be achieved by adding a load pressure feedback mechanism on the spool of the servovalve, using the similarity of its dynamics structure to that of a flexible joint robot. Finally, we show that the derived backdrivable servovalve dynamics is actually the dynamics of the pressure control valve, and we report experiments that verify the analyzed advantages of backdrivability.

Due to the confined intraocular space and physical constraints in tool manipulation, snake-like robots have a significant potential for use in retinal microsurgery. By enhancing the dexterity at the tool tip, not only the operable space on the retina can be enlarged, but also the delicate target tissues can be reached at an optimal angle minimizing the damage and making the operation much easier. In this study, we present an improved version of our earlier integrated intraocular snake (IRIS) robot, and combine it with another robotic assistant: the cooperatively controlled Steady-Hand Eye Robot (SHER). SHER is used to drive IRIS close to the retina with precision, while IRIS makes omnidirectional bends by combining its yaw and pitch motions and provides a significantly enhanced intraocular dexterity while holding the sclerotomy port fixed. For precise control of IRIS, its snake-like tip actuation has been characterized through experiments considering both a free tool tip and external loading at the tool tip. The workspace analysis showed ±45° yaw and pitch with excellent repeatability (±1°) despite the highly miniaturized articulated segment length (3 mm) and very thin shaft (0 0.9 mm). Our preliminary experiments in an artificial eye model have shown feasibility in reaching targets requiring bends up to 55° accurately.

In this paper, we present Rapid Activity Prediction Through Object-oriented Regression (RAPTOR), a scalable method for performing rapid, real-time activity recognition and prediction that achieves state-of-the-art classification accuracy on both a generic human activity dataset and two domain-specific collaborative robotics manufacturing datasets. Our approach is designed to be human-interpretable: able to provide explanations for its reasoning such that non-experts can better understand and improve its activity models. We incorporate methods to increase RAPTOR's resilience against confusion due to temporal variations, as well as against learning false correlations between features. We report full and partial trajectory classification results across three datasets and conclude by demonstrating our model's ability to provide interpretable explanations of its reasoning using outlier detection techniques.

Consider how a new robot is designed. Starting from a relative size (e.g., nanometer, centimeter, meter), the roboticist picks a morphological type (manipulator, quadruped, biped), and then uses intuition, experience, biological inspiration, or some combination of the three to select kinematic, dynamic, and geometric parameters. The designer then conducts preliminary checks to see whether the robot can satisfy its intended function: manipulation, locomotion, or both. The designer then iteratively adjusts the physical parameters and conducts checks using sample tasks, presumably until convergence is reached. This paper describes a means to automate part of this approach by combining interactive elements with powerful tools that use multi-rigid body simulation. We describe and demonstrate a virtual testing phase that first determines whether the physically situated robot would serve its intended purpose. If the robot is not capable of performing its target task, the virtual testing phase can determine which of the robot's morphological parameters should be modified in order to do so. The process keeps a human in the loop to help account for hard to quantify design aspects like appearance, quirks of the fabrication procedure (i.e., laser cutting, milling, 3D printing processes), or even expert knowledge. We intend for the described approach to be used as an interactive tool that gives a robot designer feedback on what morphological parameters are likely to limit the performance of a robot and how to modify the design to fix or offset such limitations. We demonstrate that, through simulated prototyping and testing methods, we can improve a robot design and iteratively locate morphological parameters that make efficient use of available hardware.

In this paper, elastically actuated bi-articular manipulator is developed using a novel compact planetary geared elastic actuator. The newly developed SEA, which is compact and has little backlash, can achieve high force control performance. The configuration of the developed manipulator is designed based on the bi-articular muscle coordination, which can simplify the kinematics of the manipulator. The SEA and bi-articular actuator configuration of the developed manipulator can achieve high performance force control in the workspace, and thus can be utilized for various applications related to the service robot. The mechanism and control algorithm of the novel SEA, and the novel kinematics and dynamics of the developed manipulator are introduced in this paper. Impedance controller is designed based on the kinematics and dynamics of the manipulator and its performance is verified through experiments.

In this work, we present an algorithm that simultaneously searches for a high quality fingertip grasp and a collision-free path for a robot hand-arm system to achieve it. The algorithm combines a bidirectional sampling-based motion planning approach with a hierarchical contact optimization process. Rather than tackling these problems in a decoupled manner, the grasp optimization is guided by the proximity to collision-free configurations explored by the motion planner. We implemented the algorithm for a 13-DoF manipulator and show that it is capable of efficiently planning reachable high quality grasps in cluttered environments. Further, we show that our algorithm outperforms a decoupled integration in terms of planning runtime.

As autonomous, mobile robots are increasingly entering our everyday lives and the tasks they are to perform are getting continuously more complex and versatile, instructing robots by means of natural-language commands becomes more and more important. Such instructions, stated by humans and originally intended for human use, are typically formulated very vaguely and lack critical information about how to perform particular actions. Probabilistic relational models have shown promise in filling in missing information pieces that have been omitted in such instructions. However, the enormous size of these models and the computational expense in learning and reasoning often impedes their practical applicability to real-world domains. In this work, we propose a novel instance-based learning approach towards building up knowledge bases for instruction completion, which combines probabilistic methods with semantic analogical reasoning. Probabilistic reasoning is employed to build up a knowledge base of natural-language instruction sheets, while instruction completion can be achieved through fast database queries. We showcase the scalabilty of our approach by building up a KB of more than 100,000 instruction steps that have been mined from the wikihow.com web site and which are publicly accessible from within the Prac [21] natural-language interpreter.

Pose Graph Optimization (PGO) is the de facto choice to solve the trajectory of an agent in Simultaneous Localization and Mapping (SLAM). The Maximum Likelihood Estimation (MLE) for PGO is a non-convex problem for which no known technique is able to guarantee a globally optimal solution under general conditions. In recent years, Lagrangian duality has proved suitable to provide good, frequently tight relaxations of the hard PGO problem through convex Semidefinite Programming (SDP). In this work, we build from the state-of-the-art Lagrangian relaxation [1] and contribute a complete recovery procedure that, given the (tractable) optimal solution of the relaxation, provides either the optimal MLE solution if the relaxation is tight, or a remarkably good feasible guess if the relaxation is non-tight, which occurs in specially challenging PGO problems (very noisy observations, low graph connectivity, etc.). In the latter case, when used for initialization of local iterative methods, our approach outperforms other state-of-the-art approaches converging to better solutions. We support our claims with extensive experiments.

A big challenge in environmental monitoring is the spatiotemporal variation of the phenomena to be observed. To enable persistent sensing and estimation in such a setting, it is beneficial to have a time-varying underlying environmental model. Here we present a planning and learning method that enables an autonomous marine vehicle to perform persistent ocean monitoring tasks by learning and refining an environmental model. To alleviate the computational bottleneck caused by large-scale data accumulated, we propose a framework that iterates between a planning component aimed at collecting the most information-rich data, and a sparse Gaussian Process learning component where the environmental model and hyperparameters are learned online by taking advantage of only a subset of data that provides the greatest contribution. Our simulations with ground-truth ocean data shows that the proposed method is both accurate and efficient.

We introduce an information theoretic model predictive control (MPC) algorithm capable of handling complex cost criteria and general nonlinear dynamics. The generality of the approach makes it possible to use multi-layer neural networks as dynamics models, which we incorporate into our MPC algorithm in order to solve model-based reinforcement learning tasks. We test the algorithm in simulation on a cart-pole swing up and quadrotor navigation task, as well as on actual hardware in an aggressive driving task. Empirical results demonstrate that the algorithm is capable of achieving a high level of performance and does so only utilizing data collected from the system.

Compliance is an important feature of human locomotion, where many studies have focused on the identification of stiffness parameters in order to build new machines, and stiffness is often assumed to be constant. However, it has been shown that during human walking, stiffness profiles are not constant in time and have high modulations in particular when changing walking phases. In order to understand how the constant stiffness assumption influences the walking gait and how much modulation is needed in walking mechanisms to reproduce natural gaits, in this work we focus on the analysis of joint and bi-articular stiffness profiles modulation and the influence of these modulations on the walking motions in different walking environments, i.e. level ground, slope and stairs. The analysis is carried out on periodic steps from motion capture data. mapped on a 14 segments human model with spring damper systems.

Most of the wearable robots today assist their users by acting in parallel or in series to their natural limbs. We propose a different approach to wearable robotics, consisting of devices that provide users with additional, independent robotic limbs. We present a wearable robot prototype that can achieve these goals with an extremely light weight apparatus. In order to control additional robotic limbs as if they were part of the user's body, we need voluntary signals that are independent of natural limb motions and comfortable to measure. One suitable solution - explored in this study - is the use of muscle activation signals generated by the torso. We hypothesize that a human is competent to move the extra limbs voluntarily and independently without interfering with the natural arms and legs. We developed a wearable suit to measure these signals, and we tested three possible real-time control strategies linking torso muscle contraction to the motions of two simulated extra limbs. The experimental data show that the velocity control strategy yields the highest motion accuracy, minimum muscular effort, maximum independence from the natural limbs and the fastest learning rate. This control strategy has then been applied to the control of the physical robot prototype, worn by human subjects. All of the subjects achieved accurate (normalized tracking error <; 0.5), independent (normalized natural arm motions <; 0.15) control of the extra limbs.

We consider object recognition in the context of lifelong learning, where a robotic agent learns to discriminate between a growing number of object classes as it accumulates experience about the environment. We propose an incremental variant of the Regularized Least Squares for Classification (RLSC) algorithm, and exploit its structure to seamlessly add new classes to the learned model. The presented algorithm addresses the problem of having an unbalanced proportion of training examples per class, which occurs when new objects are presented to the system for the first time. We evaluate our algorithm on both a machine learning benchmark dataset and two challenging object recognition tasks in a robotic setting. Empirical evidence shows that our approach achieves comparable or higher classification performance than its batch counterpart when classes are unbalanced, while being significantly faster.

We propose an alternative to the common approaches to the topological segmentation in structured or unstructured environments, Contour-Based Segmentation. It is faster and equally accurate, without the need of fine tuning parameters or heuristics. During robotic exploration, we propose an incremental version that reduces the processing time by reusing the previous segmentation. Tests demonstrate the velocity and quality in room segmentation in both batch and incremental mode. Tests also demonstrate the incremental version outperforms the state of the art in incremental topological segmentation.

Mechanics-based formulations of concentric tube robots incorporate tube bending and twisting, but do not include other phenomena that could model observed hysteretic behavior in which tube configurations reached by rotating tubes in different directions achieve different tip positions. As a step toward incorporating hysteretic tube-on-tube friction, this paper derives a model that enables computation of the contact forces applied by the tubes on each other along their lengths. To do so, it is necessary to include the small, but finite clearances between the tubes. Recasting the constrained energy minimization problem as its dual problem enables numerically efficient solution for the clearance-constrained centerlines of each tube as well as their contact forces. These variables are investigated through numerical examples and it is shown that, even without considering friction, the assumption of zero clearance can introduce tip position errors of several millimeters for clinically relevant robot lengths.

Convolutional neural networks (CNN) are a deep learning technique that has achieved state-of-the-art prediction performance in computer vision and robotics, but assume the input data can be formatted as an image or video (e.g. predicting a robot grasping location given RGB-D image input). This paper considers the problem of augmenting a traditional CNN for handling image-like input (called main-channel input) with additional, highly predictive, non-image-like input (called side-channel input). An example of such a task would be to predict whether a robot path is collision-free given an occupancy grid of the environment and the path's start and goal configurations; the occupancy grid is the main-channel and the start and goal are the side-channel. This paper presents several candidate network architectures for doing so. Empirical tests on robot collision prediction and control problems compare the proposed architectures in terms of learning speed, memory usage, learning capacity, and susceptibility to overfitting.

Particle tracking has emerged as a powerful tool for investigating the swarm control of microrobots and the dynamic biological processes in the life sciences. In seed plants, pollen tubes, a part of the male gametophyte, are excellent models for understanding plant growth and cellular behavior, because vesicle motion within pollen tubes reveals important information about vesicle function and interactions. Conventional vesicle tracking is based on spatiotemporal image analysis, which requires high-quality images and vesicles with constant velocity. For in vivo tracking, vesicles may disappear in some frames, and image sequences may have spatial and temporal distortions, which hamper vesicle tracking for broader applications. In this paper, we studied intracellular motion during pollen tube growth with an optical flow method. Streaming images from confocal and optical microscopes were recorded to study the intracellular motion of vesicles of different size. Local motion for each vesicle was detected using a local displacement vector field. The displacement from two adjacent frames was then calculated. The flow field shows information such as the dynamics of vesicle secretion, endocytosis, exocytosis, and cytoskeletal stability. Vesicles from different regions inside the tube were tracked simultaneously with a Kanade-Lucas-Tomasi (KLT) feature matching algorithm. The spatial and temporal characteristics of intracellular vesicles were evaluated. The proposed methods can be of great use for studying the dynamics of fluorescently tagged particles in biological systems.

One of the risks of the endonasal approach to skull base surgery is inadvertent damage to one of the two carotid arteries that are located behind the bone being drilled. Photoacoustic imaging, which combines a pulsed laser with an ultrasound receiver probe, has been shown to be able to image blood vessels behind bone. We therefore integrated a photoacoustic imaging system with a telerobotic system, where the pulsed laser is delivered via an optical fiber attached to the drill held by one robot arm and the ultrasound receiver is positioned, at some distance from the drilling site, by another robot arm. This paper describes a new method for accurately determining the safe region for drilling, which is defined by the center-line between the two carotid arteries, and presents the first phantom experiments with this system. The results show that the system can determine the center point with an accuracy better than 2 mm, which suggests that it may be sufficient for clinical scenarios where the two carotid arteries can be within 8 mm of each other.