ICRA 2018 论文列表

Planar object tracking is an actively studied problem in vision-based robotic applications. While several benchmarks have been constructed for evaluating state-of-the-art algorithms, there is a lack of video sequences captured in the wild rather than in constrained laboratory environment. In this paper, we present a carefully designed planar object tracking benchmark containing 210 videos of 30 planar objects sampled in the natural environment. In particular, for each object, we shoot seven videos involving various challenging factors, namely scale change, rotation, perspective distortion, motion blur, occlusion, out-of-view, and unconstrained. The ground truth is carefully annotated semi-manually to ensure the quality. Moreover, eleven state-of-the-art algorithms are evaluated on the benchmark using two evaluation metrics, with detailed analysis provided for the evaluation results. We expect the proposed benchmark to benefit future studies on planar object tracking.

Adaptive sampling has been shown to be an effective method for modeling environmental fields, such as algae concentrations in the ocean. In adaptive sampling, a robot adapts its sampling trajectory based on data that it is collecting. This data is often aggregated into models, using techniques such as Gaussian Process (G P) regression. The (hyper-)parameters for these models need to be manually set or, ideally, estimated from data. For GP regression, hyperparameters are typically estimated using prior data. This paper addresses the case where initial hyperparameters need to be estimated, but no prior data is available. Without prior data or accurately pre-defined hyperparameters, adaptive sampling techniques may fail, because there is no good model to base path planning decisions on. One method of gathering data is to perform a pilot survey. This survey needs to select informative samples for initiating the model, but without having a model to determine where best to sample. In this work, we evaluate four pilot surveys, which use a softmax function on the distance between waypoints and previously sampled data for waypoint selection. Simulation results show that pilot surveys that maximize waypoint spread over randomization lead to more stable estimation of GP hyperparameters, and create accurate models more quickly.

We propose a novel formulation of robotic pick and place as a deep reinforcement learning (RL) problem. Whereas most deep RL approaches to robotic manipulation frame the problem in terms of low level states and actions, we propose a more abstract formulation. In this formulation, actions are target reach poses for the hand and states are a history of such reaches. We show this approach can solve a challenging class of pick-place and regrasping problems where the exact geometry of the objects to be handled is unknown. The only information our method requires is: 1) the sensor perception available to the robot at test time; 2) prior knowledge of the general class of objects for which the system was trained. We evaluate our method using objects belonging to two different categories, mugs and bottles, both in simulation and on real hardware. Results show a major improvement relative to a shape primitives baseline.

Interactive perception exploits the correlation between forceful interactions and changes in the observed signals to extract task-relevant information from the sensor stream. Finding the most informative interactions to perceive complex objects, like articulated mechanisms, is challenging because the outcome of the interaction is difficult to predict. We propose a method to select the most informative action while deriving a model of articulated mechanisms that includes kinematic, geometric, and dynamic properties. Our method addresses the complexity of the action selection task based on two insights. First, we show that for a class of interactive perception methods, information gain can be approximated by the amount of motion induced in the mechanism. Second, we resort to physics simulations grounded in the real-world through interactive perception to predict possible action outcomes. Our method enables the robot to autonomously select actions for interactive perception that reveal most information, given the current knowledge of the world. This leads to improved perception and more accurate world models, finally enabling robust manipulation.

Robot-based sensitive assembly is a recent and growing trend in robotics. Force-controlled robots are expected to interact with an unknown environment using solely force feedback information. In general, the exact contact is difficult to predict due to various impact factors, such as the dynamics of the interaction, work-piece stiffness and geometry, the robot's configuration, and the efficiency of the control algorithm. Currently, there is no general indicator for evaluating the performance of a force-controlled robot. This work presents a concept of such a performance indicator. In order to test the proposed concept for comparison, an experimental setup is presented that simulates a contour-following task under force control. This setup is used to test two robots with different force-controllers and control principles, namely direct force and impedance control. The results indicate good applicability of the proposed performance indicator to benchmark force-controlled robots, and this is extensively discussed.

To reach a given destination safely and accurately, a micro aerial vehicle needs to be able to avoid obstacles and minimize its state estimation uncertainty at the same time. To achieve this goal, we propose a perception-aware receding horizon approach. In our method, a single forward-looking camera is used for state estimation and mapping. Using the information from the monocular state estimation and mapping system, we generate a library of candidate trajectories and evaluate them in terms of perception quality, collision probability, and distance to the goal. The best trajectory to execute is then selected as the one that maximizes a reward function based on these three metrics. To the best of our knowledge, this is the first work that integrates active vision within a receding horizon navigation framework for a goal reaching task. We demonstrate by simulation and real-world experiments on an actual quadrotor that our active approach leads to improved state estimation accuracy in a goal-reaching task when compared to a purely-reactive navigation system, especially in difficult scenes (e.g., weak texture).

We present a system enabling a modular robot to autonomously build structures in order to accomplish high-level tasks. Building structures allows the robot to surmount large obstacles, expanding the set of tasks it can perform. This addresses a common weakness of modular robot systems, which often struggle to traverse large obstacles. This paper presents the hardware, perception, and planning tools that comprise our system. An environment characterization algorithm identifies features in the environment that can be augmented to create a path between two disconnected regions of the environment. Specially-designed building blocks enable the robot to create structures that can augment the environment to make obstacles traversable. A high-level planner reasons about the task, robot locomotion capabilities, and environment to decide if and where to augment the environment in order to perform the desired task. We validate our system in hardware experiments.

Typical robot force control architectures have a positive force feedback loop to decouple robot dynamics from contact dynamics. Due to the noisy profile of force measurements, it is common to filter force signals by low pass filters. This paper shows that, when force feedback is filtered, robot and environment dynamics are no longer decoupled, affecting force control performance. Additionally, the perceived stiffness from the force control perspective, is correlated with the robot effective mass. To cope with this issue, a force based stiffness estimation strategy that also includes the inertial properties (effective mass) in the estimation algorithm is proposed, allowing to adapt control gains based on the robot effective mass. In this way, the perceived stiffness can be seen as a control optimization parameter, rather than a well defined physical property. Simulation and experimental results with a 1-DoF robot and 7-DoF manipulator, respectively, validate the estimation strategy, showing better force control results with the perceived stiffness in the control loop, as compared to the real environment stiffness.

Accurate traffic participant prediction is the prerequisite for collision avoidance of autonomous vehicles. In this work, we propose to predict pedestrians using goal-directed planning. For this, we infer a mixture density function for possible destinations. We use these destinations as the goal states of a planning stage that performs motion prediction based on common behavior patterns. The patterns are learned by a fully convolutional network operating on maps of the environment. We show that this entire system can be modeled as one monolithic neural network and trained via inverse reinforcement learning. Experimental validation on real world data shows the system's ability to predict both, destinations and trajectories accurately.

Pedestrian detection in fish-eye images is always an important problem in advanced driver assistance systems (ADAS). In conventional methods, pedestrian detectors will be trained using fish-eye images. But it is hard to collect and label enough fish-eye images manually. Therefore, a new strategy for training fish-eye pedestrian detectors using images from normal pedestrian datasets is proposed in this work. Concretely, Fish-eye Spatial Transformer Network (FSTN) is designed to generate pedestrian features in fish-eye images. FSTN aims to simulate distorted pedestrian features on the feature maps. Then the entire network is trained via adversary. FSTN is trained to generate examples which are difficult for pedestrian detectors to classify. So that the detectors are more robust to the deformation. FSTN can be embedded into state-of-the-art detectors easily. And the entire pedestrian detector, where the FSTN embedded, can be trained end to end via adversary. Moreover, experiments on ETH and KITTI pedestrian datasets show the slight accuracy improvement of pedestrian detection in fish-eye images using adversarial network compared with conventional methods.

In this paper, the problem associated with accurate control of a two-wheel steering mobile robot following a path is addressed thanks to a backstepping control strategy. This approach involves an observer to estimate the grip conditions, based on previous work, and the proposed control algorithm for the front axle. Since the significant parameters of the grip conditions are available from the observer, namely the sideslip angles and the cornering stiffnesses, it is then suitable to include them into an algorithm to control mobile robots and obtain a more accurate path tracking. This is made possible by gathering into a single backstepping approach both kinematic and dynamic models. This new point of view permits to take account of both kinematic and dynamic behaviors and grip parameters in the control law. The proposed approach is experimentally evaluated at different speeds and compared with two other state-of-the-art path tracking algorithms and evaluated for several values of lateral deviations.

Path clustering has found many applications in recent years. Common approaches to this problem use aggregates of the distances between points to provide a measure of dissimilarity between paths which do not satisfy the triangle inequality. Furthermore, they do not take into account the topology of the space where the paths are embedded. To tackle this, we extend previous work in path clustering with relative homology, by employing minimum homology area as a measure of distance between homologous paths in a triangulated mesh. Further, we show that the resulting distance satisfies the triangle inequality, and how we can exploit the properties of homology to reduce the amount of pairwise distance calculations necessary to cluster a set of paths. We further compare the output of our algorithm with that of DTW on a toy dataset of paths, as well as on a dataset of real-world paths.

Soft robotic actuators are highly elastic bodies that oscillate drastically once excited. This oscillation is undesirable in many applications. So far, very little studies on soft actuator damping have been reported. In this paper, we report a simple and effective vibration damping method based on passive and active particle damping. Experimental studies on the effectiveness of particle damping have been conducted. It is found that active particle damping is more effective than passive damping, nevertheless, active particle damping demands a more complicated design with extra energy source and control. Since particles are discrete matters, they can be seamless integrated into soft actuator design with only minor influence of soft actuator's compliance and softness.

Untethered mobile microgrippers exhibit flexibility and agility in small and constrained environments as precise and accurate robotic end-effectors, with promising potential applications in cell manipulation and microassembly. Here, we propose the first scheme to independently and simultaneously position two microgrippers on a horizontal plane for parallel targeted cargo delivery using a single global input. The separation and orientation of the two-microgripper pair are modulated by the local magnetic interactions between the two microgrippers, which are governed by a global magnetic field. The microgripper action of grasping or releasing cargoes is fully controlled by the global magnetic field without requiring additional thermal, chemical, or other stimuli. Thus, the proposed strategy only requires a single input, i.e., a global magnetic field, to control two microgrippers and therefore is simple to implement and fast-acting. As a demonstration, two microgrippers are maneuvered by a global magnetic field to pick up two cargoes and deliver them to their respective destinations. The parallel operation of two microgrippers can potentially double the overall throughput and enable the tasks that require team cooperations. The two 3D microgrippers configuration is intuitive in teleoperations, since it imitates the two-hand case of human beings.

This paper reports on a method for robust selection of inter-map loop closures in multi-robot simultaneous localization and mapping (SLAM). Existing robust SLAM methods assume a good initialization or an “odometry backbone” to classify inlier and outlier loop closures. In the multi-robot case, these assumptions do not always hold. This paper presents an algorithm called Pairwise Consistency Maximization (PCM) that estimates the largest pairwise internally consistent set of measurements. Finding the largest pairwise internally consistent set can be transformed into an instance of the maximum clique problem from graph theory, and by leveraging the associated literature it can be solved in realtime. This paper evaluates how well PCM approximates the combinatorial gold standard using simulated data. It also evaluates the performance of PCM on synthetic and real-world data sets in comparison with DCS, SCGP, and RANSAC, and shows that PCM significantly outperforms these methods.

We present PRM-RL, a hierarchical method for long-range navigation task completion that combines sampling-based path planning with reinforcement learning (RL). The RL agents learn short-range, point-to-point navigation policies that capture robot dynamics and task constraints without knowledge of the large-scale topology. Next, the sampling-based planners provide roadmaps which connect robot configurations that can be successfully navigated by the RL agent. The same RL agents are used to control the robot under the direction of the planning, enabling long-range navigation. We use the Probabilistic Roadmaps (PRMs) for the sampling-based planner. The RL agents are constructed using feature-based and deep neural net policies in continuous state and action spaces. We evaluate PRM-RL, both in simulation and on-robot, on two navigation tasks with non-trivial robot dynamics: end-to-end differential drive indoor navigation in office environments, and aerial cargo delivery in urban environments with load displacement constraints. Our results show improvement in task completion over both RL agents on their own and traditional sampling-based planners. In the indoor navigation task, PRM-RL successfully completes up to 215 m long trajectories under noisy sensor conditions, and the aerial cargo delivery completes flights over 1000 m without violating the task constraints in an environment 63 million times larger than used in training.

Climbing or crawling robots may be widely applied in agriculture, forestry, military, construction industry, disaster searching and rescuing, and so on. Soft robots possess better safety, flexibility, dexterity, portability, and adaption to complex environments than traditional robots. However, there are big challenges in system development, modeling and control of soft climbing robots. To address system development of a soft robot as a new type climbing robot, we present a pneumatic soft robot capable of inchworm-like locomotion, PISRob. The presented robot is composed of three soft parts in H-shaped configuration. Each part is able to perform 2D bending. While the middle part, as the main body, can bend in Ω -shape for actuation, the two end parts as legs can conduct simple bending motion for grasping or anchoring during locomotion. The system design and fabrication process of the soft robot is presented in details in this paper. A control system is developed for pneumatic actuation of the robot. Tests are carried out to get the relationship between the actuating air pressure and the step length in locomotion. Experiments of crawling on a floor and climbing on a pole are performed to verify the feasibility of development of the new soft robot and the effectiveness of the control method for the pneumatic system.

In this paper, we present the PerceptIn Robotics Vision System (PIRVS), a visual-inertial computing hardware with embedded simultaneous localization and mapping (SLAM) algorithm. The PIRVS hardware is equipped with a multi-core processor, a global-shutter stereo camera, and an IMU with precise hardware synchronization. The PIRVS software features a flexible sensor fusion approach to not only tightly integrate visual measurements with inertial measurements and also to loosely couple with additional sensor modalities. It runs in real-time on both PC and the PIRVS hardware. We perform a thorough evaluation of the proposed system using multiple public visual-inertial datasets. Experimental results demonstrate that our system reaches comparable accuracy of state-of-the-art visual-inertial algorithms on PC, while being more efficient on the PIRVS hardware.

Exploration in environments with sparse rewards has been a persistent problem in reinforcement learning (RL). Many tasks are natural to specify with a sparse reward, and manually shaping a reward function can result in suboptimal performance. However, finding a non-zero reward is exponentially more difficult with increasing task horizon or action dimensionality. This puts many real-world tasks out of practical reach of RL methods. In this work, we use demonstrations to overcome the exploration problem and successfully learn to perform long-horizon, multi-step robotics tasks with continuous control such as stacking blocks with a robot arm. Our method, which builds on top of Deep Deterministic Policy Gradients and Hindsight Experience Replay, provides an order of magnitude of speedup over RL on simulated robotics tasks. It is simple to implement and makes only the additional assumption that we can collect a small set of demonstrations. Furthermore, our method is able to solve tasks not solvable by either RL or behavior cloning alone, and often ends up outperforming the demonstrator policy.