ICRA 2022 论文列表

This paper solves the tracking control problem for a quadrotor system under the tasks of large-angle rotation and high-speed trajectory tracking. A quadrotor dynamic model is presented taking both disturbances and drag force into account. A reachability control strategy is developed for a quadrotor to track the planned attitude and position. Outdoor experiments of a circle trajectory tracking at different flight speeds validate the robustness of the proposed method. A task of flip is demonstrated to verify the effectiveness of the proposed controller under a large tilt angle.

Highly accurate and robust relocalization or localization initialization ability is of great importance for autonomous vehicles (AVs). Traditional GNSS-based methods are not reliable enough in occlusion and multipath conditions. In this paper we propose a novel long-term semantic relocalization algorithm based on HD map and semantic features which are compact in representation. Semantic features appear widely on urban roads, and are robust to illumination, weather, view-point and appearance changes. Repeated structures, missed and false detections make data association (DA) highly ambiguous. To this end, a robust semantic feature matching method based on a new local semantic descriptor which encodes the spatial and normal relationship between semantic features is performed. Further, we introduce an accurate, efficient, yet simple outlier removal method which works by assessing the local and global geometric consistencies and temporal consistency of semantic matching pairs. The experimental results on our urban dataset demonstrate that our approach performs better in accuracy and robustness compared with the current state-of-the-art methods.

Long-term 3D map management is a fundamental capability required by a robot to reliably navigate in the non-stationary real-world. This paper develops open-source, modular, and readily available LiDAR-based lifelong mapping for urban sites. This is achieved by dividing the problem into successive subproblems: multi-session SLAM (MSS), high/low dynamic change detection, and positive/negative change management. The proposed method leverages MSS and minimizes potential trajectory error; thus, a manual or good initial alignment is not required for change detection. Our change management scheme preserves efficacy in both memory and computation costs, providing automatic object segregation from a large-scale point cloud map. We verify the framework's reliability and applicability even under permanent year-level variation, through extensive real-world experiments with multiple temporal gaps (from day to year).

In this paper, we present a low-latency odometry system designed for spinning lidars. Many existing lidar odometry methods wait for an entire sweep from the lidar before processing the data. This introduces a large delay between the first laser firing and its pose estimate. To reduce this latency, we treat the spinning lidar as a streaming sensor and process packets as they arrive. This effectively distributes expensive operations across time, resulting in a very fast and lightweight system with a much higher throughput and lower latency. Our open source implementation is available at https://github.com/versatran01/llol.

While deep learning has enabled significant progress in designing general purpose robot grasping systems, there remain objects which still pose challenges for these systems. Recent work on Exploratory Grasping has formalized the problem of systematically exploring grasps on these adversarial objects and explored a multi-armed bandit model for identifying high-quality grasps on each object stable pose. However, these systems are still limited to exploring a small number or grasps on each object. We present Learned Efficient Grasp Sets (LEGS), an algorithm that efficiently explores thousands of possible grasps by maintaining small active sets of promising grasps and determining when it can stop exploring the object with high confidence. Experiments suggest that LEGS can identify a high-quality grasp more efficiently than prior algorithms which do not use active sets. In simulation experiments, we measure the gap between the success probability of the best grasp identified by LEGS, baselines, and the most-robust grasp (verified ground truth). After 3000 exploration steps, LEGS outperforms baseline algorithms on 10/14 and 25/39 objects on the Dex-Net Adversarial and EGAD! datasets respectively. We then evaluate LEGS in physical experiments; trials on 3 challenging objects suggest that LEGS converges to high-performing grasps significantly faster than baselines. See https://sites.google.com/view/LEGS-exp-grasping for supplemental material and videos.

Multi-LiDAR system is an important part of V2X (Vehicle to Everything) to enhance the perception information for unmanned vehicles. To fuse the information from multiple 3D LiDARs, accurate extrinsic calibration between the LiDARs is essential. However, the existing multi-LiDAR calibration methods mainly focus on short baseline scenarios, where multiple LiDARs are closely mounted on a single platform (e.g., an unmanned vehicle). Besides, most methods typically use a planar target for calibration. Some of the methods require the motion of the multi-LiDAR system. The above conditions severely limit the application of these methods to V2X, where LiDARs are non-movable, the baseline and viewpoint difference between the LiDARs can be very large. In order to meet these challenges, we propose an accurate and robust extrinsic calibration method for long baseline multi-LiDAR systems, named LB-L2L-Calib (Large Baseline LiDAR to LiDAR extrinsic Calibration). (1) We use a sphere as the calibration target for multiple LiDARs with large viewpoint difference, leveraging the viewpoint-invariance of the sphere. (2) A improved sphere detection and sphere center estimation strategy is introduced to detect and extract the sphere center from a cluttered point cloud in large-scale outdoor scenario. (3) A extrinsic parameter regression scheme is introduced. Both simulation and real experiments demonstrate that LB-L2L-Calib is highly accurate and robust. Quantitative results show that the rotation and translation error is less than 0.01m and 0.01° (in simulation, Gauss noise 0.03m, the distance and viewpoint difference between two LiDARs is more than 30m and 90°).

Disturbance mainly caused by recoil force in-evitably makes washing drone seriously deviate from the desired position, thereby reducing the cleaning efficiency. It is neces-sary to develop an effective anti-disturbance control method. Although some progresses have been made, the position error thereof is still large, rendering existing methods inapplicable in washing drone. In this paper, we propose a learning-based anti-disturbance control (LADC) method to significantly reduce the position error by combining robust nonlinear control and partial differential equation network (PDENet). Taking data noise into account, we use differential spectral normalization in the training of the PDENet. A distinguishing feature of our method is to directly learn PDENet parameters from flight logs without installing extra sensors. Experimental results indicate that the proposed method outperforms classical PD method and extended state observer (ESO) based control method with 70 % and 50 % reduced position error, respectively, and can be further applied in variable scenarios. Video: https: / / youtu.be/gNfLFAXalkI

This paper introduces an LL1adaptive control aug-mentation for geometric tracking control of quadrotors. In the proposed design, the LL 1 augmentation handles nonlinear (time-and state-dependent) uncertainties in the quadrotor dynamics without assuming or enforcing parametric structures, while the baseline geometric controller achieves stabilization of the known nonlinear model of the system dynamics. The LL1augmentation applies to both the rotational and the translational dynamics. Experimental results demonstrate that the augmented geomet-ric controller shows consistent and (on average five times) smaller trajectory tracking errors compared with the geometric controller alone when tested for different trajectories and under various types of uncertainties/disturbances.

Nonlinear dynamical effects are crucial to the operation of many agile robotic systems. Koopman-based model learning methods can capture these nonlinear dynamical system effects in higher dimensional lifted bilinear models that are amenable to optimal control. However, standard methods that lift the system state using a fixed function dictionary before model learning result in high dimensional models that are intractable for real time control. This paper presents a novel method that jointly learns a function dictionary and lifted bilinear model purely from data by incorporating the Koopman model in a neural network architecture. Nonlinear MPC design utilizing the learned model can be performed readily. We experimentally realized this method on a multirotor drone for agile trajectory tracking at low altitudes where the aerodynamic ground effect influences the system's behavior. Experimental results demonstrate that the learning-based controller achieves similar performance as a nonlinear MPC based on a nominal dynamics model in medium altitude. However, our learning-based system can reliably track trajectories in near-ground flight regimes while the nominal controller crashes due to unmodeled dynamical effects that are captured by our method.

Unmapped areas and aerodynamic disturbances render autonomous navigation with quadrotors extremely challenging. To fly safely and efficiently, trajectory planners and trackers must be able to navigate unknown environments with unpredictable aerodynamic effects in real-time. When encountering aerodynamic effects such as strong winds, most current approaches to quadrotor trajectory planning and tracking will not attempt to deviate from a determined plan, even if it is risky, in the hope that any aerodynamic disturbances can be resisted by a robust controller. This paper presents a novel systematic trajectory planning and tracking framework for autonomous quadrotors. We propose a Kinodynamic Jump Space Search (Kino-JSS) to generate a safe and efficient route in unknown environments with aerodynamic disturbances. A real-time Gaussian Process is employed to model the errors caused by aerodynamic disturbances, which we then integrate with a Model Predictive Controller to achieve efficient and accurate trajectory optimization and tracking. We demonstrate our system to improve the efficiency of trajectory generation in unknown environments by up to 75% in the cases tested, compared with recent state-of-the-art. We also show that our system improves the accuracy of tracking in selected environments with unpredictable aerodynamic effects. Our implementation is available in an open source package11https://github.com/Alex-yanranwang/Imperial-KinoJGM.

Kinematics and instantaneous kinematics are fundamental in many robotic tasks, such as positioning and collision avoidance. Existing learning methods mainly concern a single robot, and small-scale networks are sufficient for considerable approximation accuracy. A question is: Can we learn a kinematics model that can generalize to various robots rather than a single robot? This paper studies the kinematics learning of massive heterogeneous serial robots and the transfer of these general models to reality. We generate a dataset by randomizing dimensions, configurations, and link lengths and employ a network based on the generative pre-trained transformer to learn general kinematics mappings. We directly transfer our models for accuracy and use distillation-based transfer for computational efficiency. The results validate that our method can accurately approximate the kinematics of thousands of robot models and demonstrates generality in transfer.

Recent research has shown that guiding sampling-based planners with sampling distributions, learned from previous experiences via density estimation, can significantly decrease computation times for motion planning. We propose an algorithm that can estimate the density from the experiences of a robot with different kinematic structure, on the same task. The method allows to generalize collected data from one source manipulator to similarly designed target manipulators, significantly reducing the computation time for new queries for the target manipulator. We evaluate the algorithm in two experiments, including a constrained manipulation task with five different collaborative robots, and show that transferring information can significantly decrease planning time.

We propose a novel method that estimates the Kinematic Structure (KS) of arbitrary articulated rigid objects from event-based data. Event cameras are emerging sensors that asynchronously report brightness changes with a time resolution of microseconds, making them suitable candidates for motion-related perception. By assuming that an articulated rigid object is composed of body parts whose shape can be approximately described by a Gaussian distribution, we jointly segment the different parts by combining an adapted Bayesian inference approach and incremental event-based motion estimation. The respective KS is then generated based on the segmented parts and their respective biharmonic distance, which is estimated by building an affinity matrix of points sampled from the estimated Gaussian distributions. The method outperforms frame-based methods in sequences obtained by simulating events from video sequences and achieves a solid performance on new high-speed motions sequences, which frame-based KS estimation methods can not handle.

We propose a single-stage, category-level 6-DoF pose estimation algorithm that simultaneously detects and tracks instances of objects within a known category. Our method takes as input the previous and current frame from a monocular RGB video, as well as predictions from the previous frame, to predict the bounding cuboid and 6- DoF pose (up to scale). Internally, a deep network predicts distributions over object keypoints (vertices of the bounding cuboid) in image coordinates, after which a novel probabilistic filtering process integrates across estimates before computing the final pose using PnP. Our framework allows the system to take previous uncertainties into consideration when predicting the current frame, resulting in predictions that are more accurate and stable than single frame methods. Extensive experiments show that our method outperforms existing approaches on the challenging Objectron benchmark of annotated object videos. We also demonstrate the usability of our work in an augmented reality setting.

The use of drones in human-populated areas is increasing day by day. Such robots flying in close proximity to humans and potentially interacting with them, as in object handover or delivery, need to carefully plan their navigation considering the presence of humans. We propose a humanaware 3D reactive planner based on stochastic optimization for drone navigation. Besides considering the kinematics constraints of the drone, we propose two criteria to produce socially acceptable trajectories. The first, called discomfort, considers the unease caused to the humans spatially close to fast-moving drones. The second, called visibility, promotes the drone's visibility for humans. We demonstrate the planner's performance and adaptability in various simulated experiments.

Predicting future trajectories of road agents is a critical task for autonomous driving. Recent goal-based trajectory prediction methods, such as DenseTNT and PECNet [1], [2], have shown good performance on prediction tasks on public datasets. However, they usually require complicated goal-selection algorithms and optimization. In this work, we propose KEMP, a hierarchical end-to-end deep learning framework for trajectory prediction. At the core of our framework is keyframe-based trajectory prediction, where keyframes are representative states that trace out the general direction of the trajectory. KEMP first predicts keyframes conditioned on the road con-text, and then fills in intermediate states conditioned on the keyframes and the road context. Under our general framework, goal-conditioned methods are special cases in which the number of keyframes equal to one. Unlike goal-conditioned methods, our keyframe predictor is learned automatically and does not require hand-crafted goal-selection algorithms. We evaluate our model on public benchmarks and our model ranked 1st on Waymo Open Motion Dataset Leaderboard (as of September 1, 2021).

Modern autonomous systems are purposed for many challenging scenarios, where agents will face unexpected events and complicated tasks. The presence of disturbance noise with control command and unknown inputs can negatively impact robot performance. Previous research of joint input and state estimation separately studied the continuous and discrete cases without any prior information. This paper combines the continuous and discrete input cases into a unified theory based on the Expectation-Maximum (EM) algorithm. By introducing prior knowledge of events as the constraint, inequality optimization problems are formulated to determine a gain matrix or dynamic weights to realize an optimal input estimation with lower variance and more accurate decision-making. Finally, statistical results from experiments show that our algorithm owns 81% improvement of the variance than KF and 47% improvement than RKF in continuous space; a remarkable improvement of right decision-making probability of our input estimator in discrete space, identification ability is also analyzed by experiments.

The increasing deployment of robots in co-working scenarios with humans has revealed complex safety and efficiency challenges in the computation of the robot behavior. Movement among humans is one of the most fundamental —and yet critical—problems in this frontier. While several approaches have addressed this problem from a purely navigational point of view, the absence of a unified paradigm for communicating with humans limits their ability to prevent deadlocks and compute feasible solutions. This paper presents a joint communication and motion planning framework that selects from an arbitrary input set of robot's communication signals while computing robot motion plans. It models a human co-worker's imperfect perception of these communications using a noisy sensor model and facilitates the specification of a variety of social/workplace compliance priorities with a flexible cost function. Theoretical results and simulator-based empirical evaluations show that our approach efficiently computes motion plans and communication strategies that reduce conflicts between agents and resolve potential deadlocks.

Velocity Planning for self-driving vehicles in a complex environment is one of the most challenging tasks. It must satisfy the following three requirements: safety with regards to collisions; respect of the maximum velocity limits defined by the traffic rules; comfort of the passengers. In order to achieve these goals, the jerk and dynamic objects should be considered, however, it makes the problem as complex as a non-convex optimization problem. In this paper, we propose a linear programming (LP) based velocity planning method with jerk limit and obstacle avoidance constraints for an autonomous driving system. To confirm the efficiency of the proposed method, a comparison is made with several optimization-based approaches, and we show that our method can generate a velocity profile which satisfies the aforementioned requirements more efficiently than the compared methods. In addition, we tested our algorithm on a real vehicle at a test field to validate the effectiveness of the proposed method.