IROS 2020 论文列表

Traffic signal controllers play an essential role in today's traffic system. However, the majority of them currently is not sufficiently flexible or adaptive to generate optimal traffic schedules. In this paper we present an approach to learn policies for signal controllers using deep reinforcement learning aiming for optimized traffic flow. Our method uses a novel formulation of the reward function that simultaneously considers efficiency and equity. We furthermore present a general approach to find the bound for the proposed equity factor and we introduce the adaptive discounting approach that greatly stabilizes learning and helps to maintain a high flexibility of green light duration. The experimental evaluations on both simulated and real-world data demonstrate that our proposed algorithm achieves state-of-the-art performance (previously held by traditional non-learning methods) on a wide range of traffic situations.

We propose an efficient edge-based stereo visual odometry (VO) using multiple quadtrees created according to image gradient orientations. To characterize edges, we classify them into eight orientation groups according to their image gradient directions. Using the edge groups, we construct eight quadtrees and set overlapping areas belonging to adjacent quadtrees for robust and efficient matching. For further acceleration, previously visited tree nodes are stored and reused at the next iteration to warm-start. We propose an edge culling method to extract prominent edgelets and prune redundant edges. The camera motion is estimated by minimizing point-to-edge distances within a re-weighted iterative closest points (ICP) framework, and simultaneously, 3-D structures are recovered by static and temporal stereo settings. To analyze the effects of the proposed methods, we conduct extensive simulations with various settings. Quantitative results on public datasets confirm that our approach has competitive performance with state-of-the-art stereo methods. In addition, we demonstrate the practical values of our system in author-collected modern building scenes with curved edges only.

The field of autonomous driving has grown tremendously over the past few years, along with the rapid progress in sensor technology. One of the major purposes of using sensors is to provide environment perception for vehicle understanding, learning and reasoning, and ultimately interacting with the environment. In this paper, we first introduce a multisensor platform allowing vehicle to perceive its surroundings and locate itself in a more efficient and accurate way. The platform integrates eleven heterogeneous sensors including various cameras and lidars, a radar, an IMU (Inertial Measurement Unit), and a GPS-RTK (Global Positioning System / Real-Time Kinematic), while exploits a ROS (Robot Operating System) based software to process the sensory data. Then, we present a new dataset (https://epan-utbm.github.io/utbm_robocar_dataset/) for autonomous driving captured many new research challenges (e.g. highly dynamic environment), and especially for long-term autonomy (e.g. creating and maintaining maps), collected with our instrumented vehicle, publicly available to the community.

Deep learning, based on which many modern algorithms operate, is well known to be data-hungry. In particular, the datasets appropriate for the intended application are difficult to obtain. To cope with this situation, we introduce a new dataset called ETRI-Activity3D, focusing on the daily activities of the elderly in robot-view. The major characteristics of the new dataset are as follows: 1) practical action categories that are selected from the close observation of the daily lives of the elderly; 2) realistic data collection, which reflects the robot's working environment and service situations; and 3) a large-scale dataset that overcomes the limitations of the current 3D activity analysis benchmark datasets. The proposed dataset contains 112,620 samples including RGB videos, depth maps, and skeleton sequences. During the data acquisition, 100 subjects were asked to perform 55 daily activities. Additionally, we propose a novel network called four-stream adaptive CNN (FSA-CNN). The proposed FSA-CNN has three main properties: robustness to spatio-temporal variations, input-adaptive activation function, and extension of the conventional two-stream approach. In the experiment section, we confirmed the superiority of the proposed FSA-CNN using NTU RGB+D and ETRI-Activity3D. Further, the domain difference between both groups of age was verified experimentally. Finally, the extension of FSA-CNN to deal with the multimodal data was investigated.

Injuries, accidents, strokes, and other diseases can significantly degrade the capabilities to perform even the most simple activities in daily life. A large share of these cases involves neuromuscular diseases, which lead to severely reduced muscle function. However, even though affected people are no longer able to move their limbs, residual muscle function can still be existent. Previous work has shown that this residual muscular activity can suffice to apply an EMG-based user interface. In this paper, we introduce DLR's robotic wheelchair EDAN (EMG-controlled Daily Assistant), which is equipped with a torque-controlled, eight degree-of-freedom light-weight arm and a dexterous, five-fingered robotic hand. Using electromyography, muscular activity of the user is measured, processed and utilized to control both the wheelchair and the robotic manipulator. This EMG-based interface is enhanced with shared control functionality to allow for efficient and safe physical interaction with the environment.

Object-level data association and pose estimation play a fundamental role in semantic SLAM, which remain unsolved due to the lack of robust and accurate algorithms. In this work, we propose an ensemble data associate strategy for integrating the parametric and nonparametric statistic tests. By exploiting the nature of different statistics, our method can effectively aggregate the information of different measurements, and thus significantly improve the robustness and accuracy of data association. We then present an accurate object pose estimation framework, in which an outliers-robust centroid and scale estimation algorithm and an object pose initialization algorithm are developed to help improve the optimality of pose estimation results. Furthermore, we build a SLAM system that can generate semi-dense or lightweight object-oriented maps with a monocular camera. Extensive experiments are conducted on three publicly available datasets and a real scenario. The results show that our approach significantly outperforms state-of-the-art techniques in accuracy and robustness. The source code is available on https://github.com/yanmin-wu/EAO-SLAM.

In this work we present a Four-Wheeled Independent Drive and Steering (4WIDS) robot named AGRO and a method of controlling its orientation while airborne using wheel reaction torques. This is the first documented use of independently steerable wheels to both drive on the ground and achieve aerial attitude control when thrown. Inspired by a cat's self-righting reflex, this capability was developed to allow emergency response personnel to rapidly deploy AGRO by throwing it over walls and fences or through windows without the risk of it landing upside down. It also allows AGRO to drive off of ledges and ensure it lands on all four wheels. We have demonstrated a successful thrown deployment of AGRO.A novel parametrization and singularity analysis of 4WIDS kinematics reveals independent yaw authority with simultaneous adjustment of the ratio between roll and pitch authority. Simple PD controllers allow for stabilization of roll, pitch, and yaw. These controllers were tested in a simulation using derived dynamic equations of motion, then implemented on the AGRO prototype. An experiment comparing a controlled and non-controlled fall was conducted in which AGRO was dropped from a height of 0.85 m with an initial roll and pitch angle of 16 degrees and -23 degrees respectively. With the controller enabled, AGRO can use the reaction torque from its wheels to stabilize its orientation within 402 milliseconds.

Reliable real-time planning for robots is essential in today's rapidly expanding automated ecosystem. In such environments, traditional methods that plan by relaxing constraints become unreliable or slow-down for kinematically constrained robots. This paper describes the algorithm Dynamic Motion Planning Networks (Dynamic MPNet), an extension to Motion Planning Networks, for non-holonomic robots that address the challenge of real-time motion planning using a neural planning approach. We propose modifications to the training and planning networks that make it possible for real-time planning while improving the data efficiency of training and trained models' generalizability. We evaluate our model in simulation for planning tasks for a non-holonomic robot. We also demonstrate experimental results for an indoor navigation task using a Dubins car.

The wheelchair is the major means of transport for elderly and physically disabled people in their daily lives. However it cannot overcome architectural barriers such as curbs and stairs. In this study, we developed an inverted-pendulum-type robotic wheelchair for climbing stairs. The wheelchair has a seat slider and two rotary links between the front and rear wheels on each side. When climbing up stairs, the wheelchair rotates the rotary links while maintaining an inverted state of a movable body by controlling the position of the center of gravity using the seat slider. In previous research, we proposed the control method for climbing up stairs using the rotary links and seat slider, confirming a period of approximately 5 s to rotate the rotary links. In this paper, we propose a control method for going down stairs, and experimentally verify the control effectiveness and stability.

We suggest a procedure for dynamic parameter estimation of serial robot manipulators. Its basic idea relies on the synthesis of an optimal manipulation trajectory, which is based on properly introduced parameter aggregates to ensure a collection of numerically well-conditioned data-sets, yielding an accurate computation of parameter estimates. The optimal trajectory itself is computed by using a memetic algorithm, which represents a metaheuristic combination of genetic and gradient based algorithms. The algorithm is experimentally verified by estimating the parameters of the UR5 robot by Universal Robots.

The detection and tracking of dynamic traffic participants (e.g., pedestrians, cars, and bicyclists) plays an important role in reliable decision-making and intelligent navigation for autonomous vehicles. However, due to the rapid movement of the target, most current vision-based tracking methods, which perform tracking in the image domain or invoke 3D information in parts of their pipeline, have real-life limitations such as lack of the ability to recover tracking after the target is lost. In this work, we overcome such limitations and propose a complete system for dynamic object tracking in 3D space that combines: (1) a 3D position tracking algorithm based on monocular camera and LIDAR for the dynamic object; (2) a re-tracking mechanism (RTM) that restore tracking when the target reappears in camera's field of view. Compared with the existing methods, each sensor in our method is capable of performing its role to preserve reliability, and further extending its functions through a novel multimodality fusion module. We perform experiments in the real-world self-driving environment and achieve a desired 10Hz update rate for real-time performance. Our quantitative and qualitative analysis shows that this system is reliable for dynamic object tracking purposes of self-driving cars.

In dynamic environments, performance of visual SLAM techniques can be impaired by visual features taken from moving objects. One solution is to identify those objects so that their visual features can be removed for localization and mapping. This paper presents a simple and fast pipeline that uses deep neural networks, extended Kalman filters and visual SLAM to improve both localization and mapping in dynamic environments (around 14 fps on a GTX 1080). Results on the dynamic sequences from the TUM dataset using RTAB-Map as visual SLAM suggest that the approach achieves similar localization performance compared to other state-of-the-art methods, while also providing the position of the tracked dynamic objects, a 3D map free of those dynamic objects, better loop closure detection with the whole pipeline able to run on a robot moving at moderate speed.

The feet of robots are typically used to design locomotion strategies, such as balancing, walking, and running. However, they also have great potential to perform manipulation tasks. In this paper, we propose a model predictive control (MPC) framework for a quadrupedal robot to dynamically balance on a ball and simultaneously manipulate it to follow various trajectories such as straight lines, sinusoids, circles and in-place turning. We numerically validate our controller on the Mini Cheetah robot using different gaits including trotting, bounding, and pronking on the ball.

This paper proposes a dynamic attention-based visual odometry framework (DAVO), a learning-based VO method, for estimating the ego-motion of a monocular camera. DAVO dynamically adjusts the attention weights on different semantic categories for different motion scenarios based on optical flow maps. These weighted semantic categories can then be used to generate attention maps that highlight the relative importance of different semantic regions in input frames for pose estimation. In order to examine the proposed DAVO, we perform a number of experiments on the KITTI Visual Odometry and SLAM benchmark suite to quantitatively and qualitatively inspect the impacts of the dynamically adjusted weights on the accuracy of the evaluated trajectories. Moreover, we design a set of ablation analyses to justify each of our design choices, and validate the effectiveness as well as the advantages of DAVO. Our experiments on the KITTI dataset shows that the proposed DAVO framework does provide satisfactory performance in ego-motion estimation, and is able deliver competitive performance when compared to the contemporary VO methods.

A reduced balance ability can lead to falls and critical injuries. To prevent falls, humans use reaction forces and torques generated by swinging their arms. In animals, we can find that a similar strategy is taken using tails. Inspired by these strategies, we propose an approach that utilizes a robotic appendage as a human balance supporter without assistance from environmental contact. As a proof of concept, we developed a wearable robotic appendage that has one actuated degree of freedom and rotates around the sagittal axis of the wearer. To validate the feasibility of our proposed approach, we conducted an evaluation experiment with human subjects. Controlling the robotic appendage we developed improved the subjects' balance ability and enabled the subject to withstand up to 22.8 % larger impulse disturbances on average than in the fixed appendage condition.

SLAM (Simultaneous Localization And Mapping) seeks to provide a moving agent with real-time self-localization. To achieve real-time speed, SLAM incrementally propagates position estimates. This makes SLAM fast but also makes it vulnerable to local pose estimation failures. As local pose estimation is ill-conditioned, local pose estimation failures happen regularly, making the overall SLAM system brittle. This paper attempts to correct this problem. We note that while local pose estimation is ill-conditioned, pose estimation over longer sequences is well-conditioned. Thus, local pose estimation errors eventually manifest themselves as mapping inconsistencies. When this occurs, we save the current map and activate two new SLAM threads. One processes incoming frames to create a new map and the other, recovery thread, backtracks to link new and old maps together. This creates a Dual-SLAM framework that maintains real-time performance while being robust to local pose estimation failures. Evaluation on benchmark datasets shows Dual-SLAM can reduce failures by a dramatic 88%.

Magnetically actuated soft robots have recently been identified for application in medicine, due to their potential to perform minimally invasive exploration of human cavities. Magnetic solutions permit further miniaturization when compared to other actuation techniques, without loss in functionalities. Our long-term goal is to propose a novel actuation method for magnetically actuated soft robots, based on dual-arm collaborative magnetic manipulation. A fundamental step in this direction is to show that this actuation method is capable of controlling up to 8 coincident, independent Degrees of Freedom (DOFs). In present paper, we prove this concept by measuring the independent wrench components on a second pair of static permanent magnets, by means of a high resolution 6-axis load cell. The experiments show dominant activation of the desired DOFs, with mean cross-activation error of the undesired DOFs ranging from 2% to 10%.

Safe autonomous driving requires robust detection of other traffic participants. However, robust does not mean perfect, and safe systems typically minimize missed detections at the expense of a higher false positive rate. This results in conservative and yet potentially dangerous behavior such as avoiding imaginary obstacles. In the context of behavioral cloning, perceptual errors at training time can lead to learning difficulties or wrong policies, as expert demonstrations might be inconsistent with the perceived world state. In this work, we propose a behavioral cloning approach that can safely leverage imperfect perception without being conservative. Our core contribution is a novel representation of perceptual uncertainty for learning to plan. We propose a new probabilistic birds-eye-view semantic grid to encode the noisy output of object perception systems. We then leverage expert demonstrations to learn an imitative driving policy using this probabilistic representation. Using the CARLA simulator, we show that our approach can safely overcome critical false positives that would otherwise lead to catastrophic failures or conservative behavior.

For mitigating joint impact torques, researchers have reduced joint stiffness by series elastic actuators, reflected inertia by low gear ratios, and friction torque from drive-trains. However, these impact mitigation methods may impair the control performance of contact forces or may increase motor and robot mass. This paper proposes a design method for achieving a balance between impact mitigation performance and force control fidelity. We introduce an inertia-to-square-torque ratio as a new index for integrating the parameters of torque generation (motor continuous torque limits, gear ratios, etc.) and the parameters of impact mitigation (joint stiffness, reflected inertia, etc.). In the process, we make a hypothesis that a motor mass is negatively correlated with the ratio. Based on the hypothesis, we calculate a joint breakdown region of impact torques, joint stiffnesses, and motor masses. Finally, we decide the drive-train specifications of JAXON3-P and demonstrate that the proposed method provides high impact mitigation and force control capabilities through several experiments including the jumping motion of 0.3 m COG height.