IROS 2020 论文列表

This work presents a model-free nonlinear controller for an ornithopter prototype with bioinspired wings and tail. The size and power requirements have been thought to allocate a customized autopilot on board. To assess the functionality and performance of the full mechatronic design, a controller has been designed and implemented to execute a prescribed perching 2D trajectory. Although functional, its 'handmade' nature forces many imperfections that cause uncertainty that hinder its control. Therefore, the controller is based on adaptive backstepping and does not require any knowledge of the aerodynamics. The controller is able to follow a given reference in flight path angle by actuating only on the tail deflection. A novel space-dependent nonlinear guidance law is also provided to prescribe the perching trajectory. Mechatronics, guidance and control system performance is validated by conducting indoor flight tests.

In this paper, we present a new approach, AKIMap, that uses an adaptive kernel inference for dense and sharp occupancy grid representations. Our approach is based on the multivariate kernel estimation, and we propose a simple, two-stage based method that selects an adaptive bandwidth matrix for an efficient and accurate occupancy estimation. To utilize correlations of occupancy observations given sparse and non-uniform distributions of point samples, we propose to use the covariance matrix as an initial bandwidth matrix, and then optimize the bandwidth matrix by adjusting its scale in an efficient, data-driven way for on-the-fly mapping. We demonstrate that the proposed technique estimates occupancy states more accurately than state-of-the-art methods given equal-data or equal-time settings, thanks to our adaptive inference. Furthermore, we show the practical benefits of the proposed work in on-the-fly mapping and observe that our adaptive approach shows the dense as well as sharp occupancy representations in a real environment.

Multi-robot systems are widely used in environmental exploration and modeling, especially in hazardous environments. However, different types of robots are limited by different mobility, battery life, sensor type, etc. Heterogeneous robot systems are able to utilize various types of robots and provide solutions where robots are able to compensate each other with their different capabilities. In this paper, we consider the problem of sampling and modeling environmental characteristics with a heterogeneous team of robots. To utilize heterogeneity of the system while remaining computationally tractable, we propose an environmental partitioning approach that leverages various robot capabilities by forming a uniformly defined heterogeneity cost space. We combine with the mixture of Gaussian Processes model-learning framework to adaptively sample and model the environment in an efficient and scalable manner. We demonstrate our algorithm in field experiments with ground and aerial vehicles.

Several powered exoskeletons have been developed and commercialized to assist people with complete spinal cord injury. For motion control of a powered exoskeleton, a normal gait pattern is often applied as a reference. However, the physical ability of paraplegics and the degrees of freedom of powered exoskeletons are totally different from those of people without disabilities. Therefore, this paper introduces a novel gait pattern depart from the normal gait, which is proper to the paraplegics. Since a human is included, the system of the powered exoskeleton has lots of motion uncertainties that may not be perfectly predicted resulting from different physical properties of paraplegics (SCI level, muscular strength of the upper body, body parameters, inertia), actions from crutches (position and timing to put), several types of training (period, methodology), etc. Then, to find a stable and safe gait pattern adapted to the individual user, an iterative way to compensate the gait pattern is also required. In this paper, human iterative learning algorithm, which utilizes the accumulated data during walking to adjust the gait trajectories is proposed. Additionally, the effectiveness of the proposed gait pattern is verified by human walking experiments.

Local navigation is an essential ability of any mobile robot working in a real-world environment. One of the most commonly used methods for local navigation is the Dynamic Window Approach (DWA), which heavily depends on the settings of the parameters in its cost function. Since the optimal choice of the parameters depends on the environment that may significantly vary and change at any time, the parameters should be chosen dynamically in a data-driven way. To cope with this problem, we propose a novel deep convolutional neural network, which dynamically predicts these parameters considering the sensor readings. The network is trained using a state-of-the art reinforcement learning algorithm. In this way, we combine the power of data-driven learning and the dynamic model of the robot, enabling adaptation to the current environment as well as guaranteeing collision-free movement and smooth trajectories of the mobile robot. The experimental results show that the proposed method outperforms the DWA method as well as its recent extension.

In sim-to-real transfer of Reinforcement Learning (RL) policies for robot tasks, Domain Randomization (DR) is a widely used technique for improving adaptability. However, in DR there is a conflict between adaptability and training stability, and heavy DR tends to result in instability or even failure in training. To relieve this conflict, we propose a new algorithm named Domain Decomposition (DD) that decomposes the randomized domain according to environments and trains a separate RL policy for each part. This decomposition stabilizes the training of each RL policy, and as we prove theoretically, the adaptability of the overall policy can be preserved. Our simulation results verify that DD really improves stability in training while preserving ideal adaptability. Further, we complete a complex real-world vision-based patrolling task using DD, which demonstrates DD’s practicality. A video is attached as supplementary material.

We study active preference learning as a frame-work for intuitively specifying the behaviour of autonomous robots. A user chooses the preferred behaviour from a set of alternatives, from which the robot learns the user's preferences, modeled as a parameterized cost function. Previous approaches present users with alternatives that minimize the uncertainty over the parameters of the cost function. However, different parameters might lead to the same optimal behaviour; as a consequence the solution space is more structured than the parameter space. We exploit this by proposing a query selection that greedily reduces the maximum error ratio over the solution space. In simulations we demonstrate that the proposed approach outperforms other state of the art techniques in both learning efficiency and ease of queries for the user. Finally, we show that evaluating the learning based on the similarities of solutions instead of the similarities of weights allows for better predictions for different scenarios.

This paper presents a novel localisation framework based on an omnidirectional camera, targeted at outdoor urban environments. Bearing only information to persistent and easily observable high-level semantic landmarks (such as lamp-posts, street-signs and trees) are perceived using a Convolutional Neural Network (CNN). The framework utilises an information theoretic strategy to decide the best viewpoint to serve as an input to the CNN instead of the full 360° coverage offered by an omnidirectional camera, in order to leverage the advantage of having a higher field of view without compromising on performance. Environmental landmark observations are supplemented with observations to ground surface boundaries corresponding to high-level features such as manhole covers, pavement edges and lane markings extracted from a second CNN. Localisation is carried out in an Extended Kalman Filter (EKF) framework using a sparse 2D map of the environmental landmarks and Vector Distance Transform (VDT) based representation of the ground surface boundaries. This is in contrast to traditional vision only localisation systems that have to carry out Visual Odometry (VO) or Simultaneous Localisation and Mapping (SLAM), since low level features (such as SIFT, SURF, ORB) do not persist over long time frames due to radical appearance changes (illumination, occlusions etc) and dynamic objects. As the proposed framework relies on highlevel persistent semantic features of the environment, it offers an opportunity to carry out localisation on a prebuilt map, which is significantly more resource efficient and robust. Experiments using a Personal Mobility Device (PMD) driven in a representative urban environment are presented to demonstrate and evaluate the effectiveness of the proposed localiser against relevant state of the art techniques.

Learning from demonstration (LfD) is an intuitive framework allowing non-expert users to easily (re-)program robots. However, the quality and quantity of demonstrations have a great influence on the generalization performances of LfD approaches. In this paper, we introduce a novel active learning framework in order to improve the generalization capabilities of control policies. The proposed approach is based on the epistemic uncertainties of Bayesian Gaussian mixture models (BGMMs). We determine the new query point location by optimizing a closed-form information-density cost based on the quadratic Rényi entropy. Furthermore, to better represent uncertain regions and to avoid local optima problem, we propose to approximate the active learning cost with a Gaussian mixture model (GMM). We demonstrate our active learning framework in the context of a reaching task in a cluttered environment with an illustrative toy example and a real experiment with a Panda robot.

Achieving and maintaining line-of-sight (LOS) is challenging for underwater optical communication systems, especially when the underlying platforms are mobile. In this work, we propose and demonstrate an active alignment controlbased LED-communication system that uses the DC value of the communication signal as feedback for LOS maintenance. Utilizing the uni-modal nature of the dependence of the light signal strength on local angles, we propose a novel triangular exploration algorithm, that does not require the knowledge of the underlying light intensity model, to maximize the signal strength that leads to achieving and maintaining LOS. The method maintains an equilateral triangle shape in the angle space for any three consecutive exploration points, while ensuring the consistency of exploration direction with the local gradient of signal strength. The effectiveness of the approach is first evaluated in simulation by comparison with extremum-seeking control, where the proposed approach shows a significant advantage in the convergence speed. The efficacy is further demonstrated experimentally, where an underwater robot is controlled by a joystick via LED communication.

This work studies the problem of predicting the sequence of future actions for surrounding vehicles in real-world driving scenarios. To this aim, we make three main contributions. The first contribution is an automatic method to convert the trajectories recorded in real-world driving scenarios to action sequences with the help of HD maps. The method enables automatic dataset creation for this task from large-scale driving data. Our second contribution lies in applying the method to the well-known traffic agent tracking and prediction dataset Argoverse, resulting in 228,000 action sequences. Additionally, 2,245 action sequences were manually annotated for testing. The third contribution is to propose a novel action sequence prediction method by integrating past positions and velocities of the traffic agents, map information and social context into a single end-to-end trainable neural network. Our experiments prove the merit of the data creation method and the value of the created dataset - prediction performance improves consistently with the size of the dataset and shows that our action prediction method outperforms comparing models.

We consider the problem of acquiring mechanical knowledge through visual cues to help robots use objects in new situations. In this work, we propose a novel deep learning approach that allows a robot to acquire mechanical knowledge from 3D point clouds. This presents two main challenges. The first challenge is that a robot needs to infer novel objects' functions from its experience. Secondly, the robot should also need to know how to manipulate these novel objects. To solve these problems, we present a two-branch deep neural network. The first branch detects function parts from the point clouds while the second branch predicts offset poses. Fusing the results from these two branches, our approach can not only detect what functions the novel objects may have but also generate key object states which can be used to guide a robot to manipulate these objects. We show that even though most of the training samples are synthetic data, our model still learns useful features and outputs proper results. Finally, we evaluate our approach on a real robot to run a series of tasks. The experimental results show that our approach has the capability to transfer mechanical knowledge in new situations.

Collision detection is critical for safe robot operation in the presence of humans. Acoustic information originating from collisions between robots and objects provides opportunities for fast collision detection and localization; however, audio information from microphones on robot manipulators needs to be robustly differentiated from motors and external noise sources. In this paper, we present Panotti, the first system to efficiently detect and localize on-robot collisions using low-cost microphones. We present a novel algorithm that can localize the source of a collision with centimeter level accuracy and is also able to reject false detections using a robust spectral filtering scheme. Our method is scalable, easy to deploy, and enables safe and efficient control for robot manipulator applications. We implement and demonstrate a prototype that consists of 8 miniature microphones on a 7 degree of freedom (DOF) manipulator to validate our design. Extensive experiments show that Panotti realizes near perfect on-robot true positive collision detection rate with almost zero false detections even in high noise environments. In terms of accuracy, it achieves an average localization error of less than 3.8 cm under various experimental settings.

Autonomous racing provides the opportunity to test safety-critical perception pipelines at their limit. This paper describes the practical challenges and solutions to applying state-of-the-art computer vision algorithms to build a low-latency, high-accuracy perception system for DUT18 Driverless (DUT18D), a 4WD electric race car with podium finishes at all Formula Driverless competitions for which it raced. The key components of DUT18D include YOLOv3-based object detection, pose estimation, and time synchronization on its dual stereovision/monovision camera setup. We highlight modifications required to adapt perception CNNs to racing domains, improvements to loss functions used for pose estimation, and methodologies for sub-microsecond camera synchronization among other improvements. We perform a thorough experimental evaluation of the system, demonstrating its accuracy and low-latency in real-world racing scenarios.

Hyper-redundant manipulators driven by cables are used in minimally invasive surgery because of their flexibility and small diameters. In particular, manipulators composed of many rigid links and joints have the advantages of high stiffness and payload. However, these manipulators have difficulty in estimating their positions and shapes using calculations based only on the kinematics model that assumes all joint angles are equal. In this paper, we present a method for estimating the position and shape of the rolling joint in hyper-redundant manipulators by minimizing the joint moments. This allows the determination the equilibrium position of all segments of the rolling joint, and therefore an estimation of its shape. We experimentally determine the position and shape of a prototype of the rolling joint and compare them to a simulation of our method. The maximum error between the simulation and the experimental results is 4.13 mm, which is a 77.22% improvement over the kinematic model that calculates the same joint angle. This verifies that our method accurately estimates the position and shape of the rolling joint.

We propose a novel teach-and-repeat navigation system, SSM-Nav, which is based on the output of the recently introduced SSM visual place recognition methodology. During the teach phase, a teleoperated wheeled robot stores in a database features of images taken along an arbitrary route. During the repeat phase or navigation, a CNN-based comparison of each captured image is performed against the database. With the help of a particle filter, the image associated with the most likely location is selected at each time and its horizontal offset with respect to the current scene used to correct the steering of the robot and to navigate. Indoor tests in our lab show a maximum error of less than 10cm and excellent robustness to perturbations such as drastic changes in illumination, lateral displacements, different starting positions, or even kidnapping. Preliminary outdoor tests on a 0.22km route show promising results, with an estimated maximum error of less than 25cm.

This paper presents the perception, mapping, and planning pipeline implemented on an autonomous race car. It was developed by the 2019 AMZ driverless team for the Formula Student Germany (FSG) 2019 driverless competition, where it won 1st place overall. The presented solution combines early fusion of camera and LiDAR data, a layered mapping approach, and a planning approach that uses Bayesian filtering to achieve high-speed driving on unknown race tracks while creating accurate maps. We benchmark the method against our team's previous solution, which won FSG 2018, and show improved accuracy when driving at the same speeds. Furthermore, the new pipeline makes it possible to reliably raise the maximum driving speed in unknown environments from 3 m/s to 12 m/s while still mapping with an acceptable RMSE of 0.29 m.

Determining a feasible path for nonholonomic mobile manipulators operating in congested environments is challenging. Sampling-based methods, especially bi-directional tree search-based approaches, are amongst the most promising candidates for quickly finding feasible paths. However, sampling uniformly when using these methods may result in high computation time. This paper introduces two techniques to accelerate the motion planning of such robots. The first one is coordinated focusing of samples for the manipulator and the mobile base based on the information from robot surroundings. The second one is a heuristic for making connections between the two search trees, which is challenging owing to the nonholonomic constraints on the mobile base. Incorporating these two techniques into the bi-directional RRT framework results in about 5x faster and 10x more successful computation of paths as compared to the baseline method.