IROS 2016 论文列表

Object tracking over image sequences plays an remarkably crucial role in several computer vision applications, interalia, automated video surveillance, unmanned aerial vehicles and 3D reconstruction. In this paper, a novel, accurate, robust and recoverable real-time feature-based tracking framework is presented. The appearance modelling consists of a local and global layer. We propose a recommended keypoint-aware (RKA) tracker, which is fast and accurate, for the former, while the latter employs support vector machine (SVM) to determine the object and background, so that the RKA tracker can be recovered under possible target losing circumstances. Furthermore, the RKA tracker converts the tracking problem into the ranking of samples which provides a score of tracking confidence. Therefore, the priority switching between the local layer and global layer dependent upon the score becomes valid. Extensive experiments have been done by strictly following the visual tracking benchmark v1.0 protocol. The results demonstrate that the proposed novel method outperforms the state-of-the-art trackers in terms of robustness, speed and accuracy.

We propose a new class of robotic device for minimally-invasive surgery that lies at the intersection of continuum, parallel, and reconfigurable robotics. This Continuum Reconfigurable Incisionless Surgical Parallel (CRISP) paradigm involves the use of multiple needle-diameter devices inserted through the skin and assembled into parallel structures inside the body. The parallel structure can be reconfigured inside the patient's body to satisfy changing task requirements such as reaching initially inaccessible locations or modifying mechanical stiffness for manipulation or palpation. Another potential advantage of the CRISP concept is that many small (needle-sized) entry points into the patient may be preferable in terms of both patient healing and cosmesis to the single (or multiple) larger ports needed to admit current surgical robots. This paper presents a mechanics-based model for CRISP forward and inverse kinematics, along with experimental validation.

Localization of a human or a robot in indoor environment prior to initial human-robot interaction is the motivation of this paper. The utilization of Wi-Fi received signal strength (RSS) by adopting fingerprinting technique is used to compute the location estimation. However, Wi-Fi RSS instability incurred by mutable channel characteristics hampers a wide-spread adoption of RSS based location fingerprinting to real world applications. To overcome RSS instability problem, we propose a new method based on the concept of “invariant RSS statistics”, where it is defined as the RSS samples collected at each calibration location under minimal random spatiotemporal disturbances. The invariant RSS statistics thus collected forms the reference pattern classes in the space formed by Wi-Fi sources. Fingerprinting is done by identifying the reference pattern class that maximally supports the RSS readings collected at an unknown location for available Wi-Fi sources. The support of RSS readings is defined here as the sum of the likelihood probabilities of individual RSS readings. Experimental results show that the proposed method provides superior performance to conventional ones with the success rate higher by 17%, the printing resolution finer by 30% and, naturally, no performance degradation in time without recalibration. Lastly, we present a design guideline that relates statistically the different levels of the random spatiotemporal disturbances inducing RSS instability to the minimum number of Wi-Fi sources required for achieving a certain class separation degree under the given number of calibration locations to be identified.

In this paper, we propose a real-time control method for skating motion of humanoid robots. There are three problems for skating motion: (1) keeping dynamic balance, (2) adequately controlling foot force to suppress slipping at the foot, (3) controlling full-body motion in real-time. For solving these problems, we propose the Skating Motion Generator and the Skating Motion Stabilizer. In the Skating Motion Generator, we separate the slip suppression from motion generation for (3). The separation enables us to generate skating motions in real-time. In the Skating Motion Stabilizer, we adjust the sole pressure distribution of each foot to solve the contradiction between (1) and (2). We show the effectiveness of the proposed controller through the experiments, in which life-sized humanoid HRP-2 pushes the ground and skates on the skateboard. Applying the proposed controller, HRP-2 could successfully accelerate and skate on the skateboard at 0.5[m/s].

Low-drift rotation estimation is a crucial part of any accurate odometry system. In this paper, we focus on the problem of 3D rotation estimation with dense depth sensors in environments that consist of piece-wise planar structures, such as corridors and office rooms. An efficient mean-shift paradigm is developed to extract and track planar modes in the surface normal vector distribution on the unit sphere. Robust and piece-wise drift-free behavior is achieved by registering the bundle of planar modes from the current frame with respect to a reference frame using a general ℓ1-norm regression scheme. We furthermore add a memory scheme to the regular birth and death of modes, which further compensates accumulated rotational drift when previously discovered modes are revisited. We discuss the robustness issue and evaluate our algorithm on both custom synthetic as well as real publicly available datasets. Our experimental results demonstrate high robustness and effectiveness of the proposed algorithm.

This paper presents a real-time image restoration method for aerial inspection images that are degraded by the appearance of a passive rotating shell of an unmanned aerial vehicle (UAV). Only images are required as the input. The method mainly consists of feature-based object detection for joints of the UAV shell, outlier rejection based on a sample consensus (SAC) approach and the geometrical model of the shell, reconstruction of the entire region that the shell appears in, and inpainting of the shell. Not only a pure algorithm for general UAVs with passive rotating shells but also an implementation specialized for a UAV that we have been developing are presented. The successfully implemented algorithm was evaluated for actual inspection flights under different conditions. The average rate of successful image restoration was 81.1% at 4.3 fps.

Real-time 3D perception is critical for localisation, mapping, path planning and obstacle avoidance for mobile robots and autonomous vehicles. For outdoor operation in real-world environments, 3D perception is often provided by sparse 3D LIDAR scanners, which provide accurate but low-density depth maps, and dense stereo approaches, which require significant computational resources for accurate results. Here, taking advantage of the complementary error characteristics of LIDAR range sensing and dense stereo, we present a probabilistic method for fusing sparse 3D LIDAR data with stereo images to provide accurate dense depth maps and uncertainty estimates in real-time. We evaluate the method on data collected from a small urban autonomous vehicle and the KITTI dataset, providing accuracy results competitive with state-of-the-art stereo approaches and credible uncertainty estimates that do not misrepresent the true errors, and demonstrate real-time operation on a range of low-power GPU systems.

In this paper, we present a novel pose estimation and obstacle avoidance approach for tendon-driven multi-segment continuum manipulators moving in dynamic environments. A novel multi-stage implementation of an Extended Kalman Filter is used to estimate the pose of every point along the manipulator's body using only the position information of each segment tip. Combined with a potential field, the overall algorithm will guide the manipulator tip to a desired target location and, at the same time, keep the manipulator body safe from collisions with obstacles. The results show that the approach works well in a real-time simulation environment that contains moving obstacles in the vicinity of the manipulator.

The article proposes a solution allowing a multicopter to generate and follow a path while taking into account the obstacles in the environment.

In this paper we propose a new vision-based obstacle avoidance strategy using the Underwater Dark Channel Prior (UDCP) that can be applied to any Unmanned Underwater Vehicle (UUV) equipped with a simple monocular camera and minimal on-board processing capabilities. For each incoming image, our method first computes a relative depth map to estimate the obstacles nearby. Then, the map is segmented and the most promising Region of Interest (RoI) is identified. Finally, an escape direction is computed within the RoI and a control action is performed accordingly to avoid the obstacles. We tested our approach on a video sequence in a natural environment and compared it against a state-of-the-art method showing better performance, specially in light changing conditions. We also provide online results on a low-cost Remotely Operated Vehicle (ROV) in a controlled environment.

With society and industry pushing for robot-assisted systems to automate cumbersome tasks, such as inspection and maintenance, a vast amount of research effort has been dedicated to relevant technologies. Right at the forefront are small Unmanned Aerial Vehicles (UAVs) equipped with onboard cameras, recently demonstrating that vision-based autonomous flights without reliance on GPS are possible, sparking great interest in a plethora of areas. Current solutions, however, still lack in portability and generality struggling to perform outside the controlled laboratory environment, with onboard robotic perception constituting the biggest impediment. Driven by the need for real-time denser scene estimation, in this work we present a dramatically cheap approach enabling estimation of the immediate surroundings of a UAV using the inertial and visual cues from a single onboard camera. Instead of following the recent trend towards dense scene reconstruction, we trade detail of reconstruction for efficiency of estimation, albeit without compromising accuracy. We also present ETHZ CAB building dataset for aerial inspection. We present results against scene ground truth obtained by a millimetre-precise laser scanner.

We present a robust real-time method which performs dense reconstruction of high quality height maps from monocular video. By representing the height map as a triangular mesh, and using efficient differentiable rendering approach, our method enables rigorous incremental probabilistic fusion of standard locally estimated depth and colour into an immediately usable dense model. We present results for the application of free space and obstacle mapping by a low-cost robot, showing that detailed maps suitable for autonomous navigation can be obtained using only a single forward-looking camera.

We present a real-time framework for planning natural and smooth grasping motions in an online manner based on the interaction with human. The proposed framework is able to change its grasp strategies agilely according to the interaction with human. Given human demonstrations, we develop a motion field graph consisting of nodes and edges where the nodes contains reference finger poses and their time derivatives, and the edges indicates the similarity between a pair of nodes. Based on the graph, a new grasping motion can be planned that adapts to the changes of the environment and the interaction. The motion field guarantees smooth motions by integrating the velocities obtained from the demonstrations. To validate the framework, we build a demo system where a human can hand a cup over to a tele-operated robot or a virtual humanoid avatar which are controlled by an another person at a remote location. Also, the virtual avatar can grasp and manipulate a cola can.

With robotic systems reaching considerable maturity in basic self-localization and environment mapping, new research avenues open up pushing for interaction of a robot with its surroundings for added autonomy. However, the transition from traditionally sparse feature-based maps to dense and accurate scene-estimation imperative for realistic manipulation is not straightforward. Moreover, achieving this level of scene perception in real-time from a computationally constrained and highly shaky and agile platform, such as a small an Unmanned Aerial Vehicle (UAV) is perhaps the most challenging scenario for perception for manipulation. Drawing inspiration from otherwise computationally constraining Computer Vision techniques, we present a system combining visual, inertial and depth information to achieve dense, local scene reconstruction of high precision in real-time. Our evaluation testbed is formed using ground-truth not only in the pose of the sensor-suite, but also the scene reconstruction using a highly accurate laser scanner, offering unprecedented comparisons of scene estimation to ground-truth using real sensor data. Given the lack of any real, ground-truth datasets for environment reconstruction, our V4RL Dense Surface Reconstruction dataset is publicly available.

Real-time contamination monitoring in health care facilities would allow medical teams to take appropriate actions to carefully enter, avoid, or decontaminate contaminated areas, reducing infection risk for themselves and their patients. In this paper, we demonstrate and evaluate the first end-to-end, real-time contamination tracking system for robotic health care support. The system models contamination of the environment and people, directs decontamination efforts of a simulated scrubber robot, and alerts users when nearing contaminated areas. We outline our transmission model design choices, as based on Ebola virus disease, and evaluate the system against the spread of a physical substance.

This paper addresses the problem of real-time, non-holonomic motion planning in environments with moving obstacles of unforeseen, arbitrary motion. An approach is introduced to smoothly switch trajectories by generating feasible non-holonomic trajectory segments on the fly as the robot moves in such an environment, extending the real-time adaptive motion planning (RAMP) approach that is used for holonomic motion. It allows efficient on-line simultaneous planning and execution of non-holonomic trajectories and enables a robot to adapt to changes in the environment while taking into account robot motion uncertainty. The effectiveness and efficiency of the method has been verified through real experiments with a mobile robot and several dynamic obstacles of unforeseen motion to the robot.

Constraint based programming provides a versatile framework for combining several different constraints into a single robot control scheme. We take advantage of the redundancy of a robot manipulator to improve the execution of a reactive tracking task, in terms of a task-dependent measure which is a weighted sum of velocity transmissions along the current directions of motion. With inspiration from recent work, we provide analytical gradients and computable weights of the task-dependent measure, which enable us to include it in a reactive constraint based programming framework, without relying on inexact numerical approximations and manually tuning weights. The proposed approach is illustrated in a set of simulations, comparing the performance with a standard constraint based programming method.

In the context of visual object search and localization, saliency maps provide an efficient way to find object candidates in images. Unlike most approaches, we propose a way to learn saliency maps directly on a robot, by exploring the environment, discovering salient objects using geometric cues, and learning their visual aspects. More importantly, we provide an autonomous exploration strategy able to drive the robot for the task of learning saliency. For that, we describe the Reinforcement Learning-Intelligent Adaptive Curiosity algorithm (RL-IAC), a mechanism based on IAC (Intelligent Adaptive Curiosity) able to guide the robot through areas of the space where learning progress is high, while minimizing the time spent to move in its environment without learning. We demonstrate first that our saliency approach is an efficient tool to generate relevant object boxes proposal in the input image and significantly outperforms the state-of-the-art EdgeBoxes algorithm. Second, we show that RL-IAC can drastically decrease the required time for learning saliency compared to random exploration.