IROS 2017 论文列表

Existing robots for multi-scale motion (MSM) are unsuitable for micro-surgery in deep surgical sites where miniaturization and traversal of often tortuous anatomical passageways is required. Also, new emerging surgical paradigms for natural orifice surgery and image-based diagnosis and guidance at the micro-scale level promise to provide accurate verification of tumor resection margins. To overcome the limitations of current robot architectures, and to enable image-based biopsy and micro-surgery in confined spaces, we present a new concept of continuum robots with equilibrium modulation (CREM). CREM robots are a modification of multi-backbone continuum robots that achieve micro-motion by using indirect actuation through modulation of their static equilibrium by change of the distribution of their cross-sectional flexural rigidity. As a first step towards modeling the micro-scale kinematics of these robots, solutions for micro-motion tracking are presented and verified to achieve tracking accuracies of better than 2μm. Preliminary evaluation of the micro-motion capabilities of a first prototype demonstrates motion resolutions at 1μm level and hysteresis of less than 10μm - despite the use of inexpensive actuators with significant backlash. Finally, a demonstration of a first effort at integrating such a robot with a custom-made optical coherence tomography (OCT) probe is presented.

We propose an approach to vision-based pose estimation using object recognition and identity. Whereas feature based scene recognition and pose estimation methods are well established as effective means for estimating motion and recognizing locations, feature-based methods depend critically on the detection of common local features from one view of a scene to another. We focus on place recognition and pose change estimation in the context of large changes in viewing position, even to the extent that no common surfaces are seen between the two views. Our approach is based on using object identities and their inter-relationship to compute pose change. An important secondary outcome of our method is that it simultaneously infers the 3D poses of objects in the scene that are used as features. Such an object-based approach is inspired by a vast literature on human perception and has the potential for great robustness, albeit at the expense of accuracy. We propose a formulation of the problem using pairwise contextual constraints and develop an efficient algorithmic solution. We validate the approach and quantify its performance using the publicly available TUM SLAM dataset [1].

Differential Dynamic Programming (DDP) has become a popular approach to performing trajectory optimization for complex, underactuated robots. However, DDP presents two practical challenges. First, the evaluation of dynamics derivatives during optimization creates a computational bottleneck, particularly in implementations that capture second-order dynamic effects. Second, constraints on the states (e.g., boundary conditions, collision constraints, etc.) require additional care since the state trajectory is implicitly defined from the inputs and dynamics. This paper addresses both of these problems by building on recent work on Unscented Dynamic Programming (UDP) - which eliminates dynamics derivative computations in DDP-to support general nonlinear state and input constraints using an augmented Lagrangian. The resulting algorithm has the same computational cost as first-order penalty-based DDP variants, but can achieve constraint satisfaction to high precision without the numerical ill-conditioning associated with penalty methods. We present results demonstrating its favorable performance on several simulated robot systems including a quadrotor and 7-DoF robot arm.

This paper introduces an approach for designing exciting trajectories for parameter identification of serial robots based on a combination of Fourier Series (FS) and Schroeder Phased Harmonic Sequence (SPHS). An initial estimation of the trajectory is designed for each link using SPHS. Afterwards, the initial trajectory enable to find the initial parameters of the FS which are fed to an optimization process that finds the optimal parameters of the FS used for identification purpose. Like this, we can take the advantages of both FS and SPHS and eliminate the disadvantages of each of them. Moreover, a comparison of results between; the proposed method, original FS, and SPHS is taking place to demonstrate the effectiveness of the new approach. In this vein, a one-way analysis of variance is conducted to compare whether there are significant improvement or not. An PA10 7DoF arm robot serves as test bed for conducting experiments. Findings shows that the optimal trajectory found through the proposed approach requires less computation time compared to original FS which can be an advantage for fast and robust identification.

In this paper, we propose a unified model for camera pose estimation and a novel strategy for pose optimization by combining points and lines in monocular visual odometry. Our proposed unified model treats point and line features equivalently, which is applicable for all the minimal cases requiring the minimum number 3 of point or/and line features and can be easily extended for various circumstances with more additional observations. The core idea is to directly retrieve all stationary points of a cost function which is minimized by the first-order optimality condition without initialization or iteration. The estimated pose is reliable due to robust geometric constraints and the reliable algebraic solver. To refine the camera pose, we propose a novel optimization strategy to minimize the unconstrained Sampson error by taking specific uncertainty for each feature into account to penalize noise more reasonably. Moreover, it is simpler than the conventional bundle adjustment by avoiding the high-dimensional parameter searching. Experimental results on simulated data and real images have sufficiently demonstrated the superiority of our proposed camera pose estimation and optimization method by comparing with state-of-the-art monocular algorithms.

Task and motion planning subject to Linear Temporal Logic (LTL) specifications in complex, dynamic environments requires efficient exploration of many possible future worlds. Model-free reinforcement learning has proven successful in a number of challenging tasks, but shows poor performance on tasks that require long-term planning. In this work, we integrate Monte Carlo Tree Search with hierarchical neural net policies trained on expressive LTL specifications. We use reinforcement learning to find deep neural networks representing both low-level control policies and task-level “option policies” that achieve high-level goals. Our combined architecture generates safe and responsive motion plans that respect the LTL constraints. We demonstrate our approach in a simulated autonomous driving setting, where a vehicle must drive down a road in traffic, avoid collisions, and navigate an intersection, all while obeying rules of the road.

This paper presents image-based navigation from an image memory using a combination of line segments and feature points. The environment is represented by a set of key images, which are acquired during a prior mapping phase that defines the path to be followed during the navigation. The switching of key images is done exploiting the common line segments and feature points between the current acquired image and the nearby key images. Based on the key images and the current image, a control law is derived for computing the rotational velocity of a mobile robot during its visual navigation. Using our approach, real-time navigation has been performed in real indoor environment with a Pioneer 3-DX equipped with an on-board perspective camera and the humanoid robot Pepper without the need of accurate mapping and localization nor of 3D reconstruction. We also show that the combination of points and lines increases the number of features that helps in robust and successful navigation especially in those regions where few points or lines can be detected and tracked/matched.

The developments of robotics inform research across a broad range of disciplines. In this paper, we will study and compare two collision selective neuron models via a vision-based autonomous micro robot. In the locusts' visual brain, two Lobula Giant Movement Detectors (LGMDs), i.e. LGMD1 and LGMD2, have been identified as looming sensitive neurons responding to rapidly expanding objects, yet with different collision selectivity. Both neurons have been modeled and successfully applied in robotic vision system for perceiving potential collisions in an efficient and reliable manner. In this research, we conduct binocular neuronal models, for the first time combining the functionalities of LGMD1 and LGMD2 neurons, in the visual modality of a ground mobile robot. The results of systematic on-line experiments demonstrated three contributions of this research: (1) The arena tests involving multiple robots verified the effectiveness and robustness of a reactive motion control strategy via integrating a bilateral pair of LGMD1 and LGMD2 models for collision detection in dynamic scenarios. (2) We pinpointed the different collision selectivity between LGMD1 and LGMD2 neuron models, which fulfill corresponding biological research. (3) The utilized micro robot may also benefit researches on other embedded vision systems as well as swarm robotics.

Policy search methods and, more broadly, reinforcement learning can enable robots to learn highly complex and general skills that may allow them to function amid the complexity and diversity of the real world. However, training a policy that generalizes well across a wide range of real-world conditions requires far greater quantity and diversity of experience than is practical to collect with a single robot. Fortunately, it is possible for multiple robots to share their experience with one another, and thereby, learn a policy collectively. In this work, we explore distributed and asynchronous policy learning as a means to achieve generalization and improved training times on challenging, real-world manipulation tasks. We propose a distributed and asynchronous version of guided policy search and use it to demonstrate collective policy learning on a vision-based door opening task using four robots. We describe how both policy learning and data collection can be conducted in parallel across multiple robots, and present a detailed empirical evaluation of our system. Our results indicate that distributed learning significantly improves training time, and that parallelizing policy learning and data collection substantially improves utilization. We also demonstrate that we can achieve substantial generalization on a challenging real-world door opening task.

A popular class of lidar-based grid mapping algorithms computes for each map cell the probability that it reflects an incident laser beam. These algorithms typically determine the map as the set of reflection probabilities that maximizes the likelihood of the underlying laser data and do not compute the full posterior distribution over all possible maps. Thereby, they discard crucial information about the confidence of the estimate. The approach presented in this paper preserves this information by determining the full map posterior. In general, this problem is hard because distributions over real-valued quantities can possess infinitely many dimensions. However, for two state-of-the-art beam-based lidar models, our approach yields closed-form map posteriors that possess only two parameters per cell. Even better, these posteriors come for free, in the sense that they use the same parameters as the traditional approaches, without the need for additional computations. An important use case for grid maps is robot localization, which we formulate as Bayesian filtering based on the closed-form map posterior rather than based on a single map. The resulting measurement likelihoods can also be expressed in closed form. In simulations and extensive real-world experiments, we show that leveraging the full map posterior improves the localization accuracy compared to approaches that use the most likely map.

This research develops an amphibious transformable robot, Claw-Wheel IV, which could move on flat ground, uneven terrain, and aquatic environment. Compared to the previous version, Claw-Wheel III, the new generation has waterproof enclosure to adapt to watery environments and paddles are installed to produce underwater propulsion force, while the transformable idea and claw locomotion mechanism design are retained. Different kinds of gaits are designed and compared to reduce the vertical fluctuation of COM of the robot and run more smoothly. Advancing in aquatic environments relies on the paddles driven along with the rotating claws while steering is achieved with velocity differential between claws on both sides. Velocities of individual claws are computed and regulated based on gyration radius and angular velocity parameters obtained in experiments, in order to follow ideal trajectories.

Classification techniques are useful for processing complex signals into labels with semantic value. For example, they can be used to interpret brain signals generated by humans corresponding to a finite set of commands for a physical device. The classifier, however, may interpret the signal as a command that is different from the intended one. This error in classification leads to poor performance in tasks where the class labels are used to learn some information or to control a physical device. We propose a computationally efficient algorithm to identify which class labels may be misclassified out of a sequence of class labels, when these labels are used in a given learning or control task. The algorithm is based on inference methods using Markov random fields. We apply the algorithm to goal-learning and tracking using brain-computer interfacing (BCI), in which signals from the brain are commonly processed using classification techniques. We demonstrate that the proposed algorithm reduces the time taken to identify the goal state in control experiments.

In this paper we propose an algorithm for stabilizing circular formations of fixed-wing UAVs with constant speeds. The algorithm is based on the idea of tracking circles with different radii in order to control the inter-vehicle phases with respect to a target circumference. We prove that the origin of the error dynamics of the inter-vehicle phases is exponentially stable. Furthermore, thanks to the stability properties of the guidance vector field that guides the vehicles, the algorithm can be extended to other closed trajectories. One of the main advantages of this approach is that the algorithm guarantees the confinement of the team in a specific area, even when communications or sensing among vehicles are lost. We show the effectiveness of the algorithm with an actual formation flight of three aircraft. The algorithm is ready to use for the general public in the open-source Paparazzi autopilot.

The Freefall Camera (FFC) is a robot designed to film skydivers in free fall and is, to the author's knowledge, the first robot to control its position and terminal velocity relative to another falling object. The robot is equipped with a vision sensor to track objects of a chosen colour. The control surfaces for free fall motion are four vertical ailerons that control planar position and orientation and two retractable flaps that control terminal velocity. To develop a robust controller, equations of motion were created and validated experimentally for movement in the vertical and planar rotational directions. The static and dynamic response of the system were then tested in a vertical wind tunnel to validate the models and use them to design and tune a PID + Filter controller for the free fall flight of the robot. Planar rotation tests showed a decrease in steady state angular velocity and dynamic response gain compared to the model and a slight increase in dynamic response phase shift. This was attributed to the effect of the turbulent air near the shell of the robot. Vertical translation tests showed a decrease in maximum terminal velocity compared to the model, likely due to the effect of air viscosity at higher Reynolds numbers. The models were used to tune a PID + Filter controller to achieve stable target tracking in the rotational and vertical degrees of freedom using image based visual servoing. Controlled flight of the FFC was then tested in a vertical wind tunnel, where the robot was able to successfully track a coloured target within 15° of the centre of the field of view.

We discuss challenges a sprawling posture legged robot faces when moving in a complex environment. These kind of robots can be used for search and rescue applications when proper sensors are used and placed correctly. Finding an adequate position for placing sensors on legged robots with a segmented spine to maximize the information retrieval is not trivial. In this paper, we talk about equipping the salamanderlike robot Pleurobot with necessary sensors to understand its challenges. Our setup gathers visual and inertial information for 3D mapping and localization. To do so, we have come up with three different experimental terrains to analyze the sensing behavior. The results are examples to understanding the behavior of quadrupedal legged robots similar to Pleurobot in search and rescue scenarios.

Technologies in autonomous vehicles have seen dramatic advances in recent years; however, it still lacks of robust perception systems for car detection. With the recent development in deep learning research, in this paper, we propose a LIDAR and vision fusion system for car detection through the deep learning framework. It consists of three major parts. The first part generates seed proposals for potential car locations in the image by taking LIDAR point cloud into account. The second part refines the location of the proposal boxes by exploring multi-layer information in the proposal network and the last part carries out the final detection task through a detection network which shares part of the layers with the proposal network. The evaluation shows that the proposed framework is able to generate high quality proposal boxes more efficiently (77.6% average recall) and detect the car at the state of the art accuracy (89.4% average precision). With further optimization of the framework structure, it has great potentials to be implemented onto the autonomous vehicle.

The magnetic particles actuated by an external magnetic field present a promising solution as drug carrier at the inner ear (cochlea). This technique allows non-invasive drug delivery with minimal hydraulic force and very high precision. In this work, magnetic prototyping platform for intra cochlear drug delivery procedure is proposed. This platform is composed by magnetic actuator mounted on a robot end-effector and an hybrid vision system. The latter consists of a camera to control the position of magnetic actuator, two microscopes to recover the position of the magnetic particles and the 3D structure of the cochlea. The hybrid vision system allows real-time tracking of magnetic particles inside the artificial cochlea. The accuracy of the drug delivery strategy depends strongly on the calibration of the vision system. In this paper a method to calibrate the hybrid vision system is proposed and validated with experimental results.

We address the problem of overapproximative short-term prediction of arm movement, for safe human-robot co-existence. A trajectory planner for a robot manipulator which formally verifies non-collision requires fast prediction of human future occupancy, which must be accurate yet still account for all possible human movement. This work presents two approaches to this: a novel, simple approach calculated directly in task space, and another approach using a kinematic model of the human based on the authors' previous work. We compare both approaches in an experimental setup with a robot manipulator. To the best of our knowledge, this is the first implementation and comparison of approaches to formally verified trajectory planning in human-robot co-working. We find that the novel approach has advantages in terms of ease and speed of computation and tightness at short time horizons and is more intuitive to extend to the whole human body; the previous approach offers advantages at longer prediction horizons. We also perform conformance checking to show that both approaches do indeed account for all relevant movement.

Underwater communication is particularly challenging for small submarine robots that have limited power and size constraints. Inspired by weakly electric fish, a novel electric current communication (termed electrocommunication) system has been developed for small underwater robots in our previous studies. However, collision problems sometimes occur during multiple robots communication because the electrocommunication network shares a common channel. In this paper, a new CSMA/CA-based electrocommunication system is presented to solve this collision problem. An efficient circuit for communication channel state (busy or idle) detection is proposed. After that, a compact Carrier Sense Multiple Access with Collision Avoidance (CSMA/CA) algorithm is introduced and finally implemented on the system to judiciously avoid collision during multiple robot communication. The effectiveness of the proposed CSMA/CA-based communication system for underwater robot groups is well verified by both simulations and experiments with three robotic fish models.