IROS 2018 论文列表

This paper describes a new architecture of deep neural networks (DNNs) for acoustic models. Training DNNs from raw speech signals will provide 1) novel features of signals, 2) normalization-free processing such as utterance-wise mean subtraction, and 3) low-latency speech recognition for robot audition. Exploiting the longer context of raw speech signals seems useful in improving recognition accuracy. However, naive use of longer contexts results in the loss of short-term patterns; thus, recognition accuracy degrades. We propose a multi-timescale feature-extraction architecture of DNNs with blocks of different time scales, which enable capturing long- and short-term patterns of speech signals. Each block consists of complex-valued networks that correspond to Fourier and filterbank transformations for analysis. Experiments showed that the proposed multi-timescale architecture reduced the word error rate by about 3% compared with those only with the longterm context. Analysis of the extracted features revealed that our architecture efficiently captured the slow and fast changes of speech features.

Recently, self-supervised approach is common for robot grasping. Although this approach improves success rate, it requires a long time to execute a number of grasp trials, and single-arm grasping is only considered. However, robots can grasp more various objects with two arms, and dual-arm robots such as humanoid robots are expected to execute dual-arm manipulation and overcome the single-arm limitation. In this paper, we introduce dual-arm grasping as another possible strategy and propose a multi-stage learning method for selective dual-arm grasping using Convolutional Neural Networks (CNN)for grasping point prediction and semantic segmentation. In the first stage, the network learns grasping points with the automatic annotation. Although a robot learns both single-arm and dual-arm grasping efficiently with the annotation, the robot may not be able to grasp it because the annotation algorithm is designed by human. Therefore, for the second stage, the robot samples various grasping points with both grasping strategies and learns how to grasp in the real world. In this stage, the robot obtains new possible grasping points and prunes unsuccessful ones for both grasping strategies through the robot experience. In the experiments in the real world, the adapted network achieved high success rate 76.7% in 90 trials. Since the network trained with no adaptation stage resulted in lower success rate 56.7%, this result also shows the network was refined with less than 250 times of grasp sampling. As an application of our method, we demonstrated that our system worked well in warehouse picking task.

This paper presents a switching strategy of a region-based shape controller for a swarm-robotic framework to overcome the traditional obstacle-avoidance problem in the virtual structure methodology. In this control approach, initially, the robots move as a group inside a circular region which we conceive to be the initial virtual structure, while preserving a specific pattern, say a triangular formation, among them. In order to avoid static/dynamic obstacles, while approaching towards the target without any prior knowledge about the environment, the virtual-circle is allowed to shrink up to a certain limit. The shrinking phenomena of the virtual circle will depend upon the number of agents within the circle and the distance between two the nearest obstacles sensed by the agents through which the swarm should be able to pass. If the situation demands, the structure may assume the shape of an ellipse of equivalent area continually throughout the path described by the swarm encapsulated within the variable structure. To achieve this, two-layer hierarchical control strategy has been proposed. Moreover, if the shape of the virtual structure changes, the formation of the swarm inside the region may also change. To make the inter-agent formation flexible inside the newly formed virtual structure, a spanning-tree-assisted-shape-matching algorithm has been employed for accommodating all the agents inside the virtual region which helps in the formation change in the agents as well. Finally, simulation results and stability analysis of the controllers are provided to demonstrate our proposed technique.

Robot learning by demonstration gives robots the ability to learn tasks which they have not been programmed to do before. The paradigm allows robots to work in a greater range of real-world applications in our daily life. However, this paradigm has traditionally been applied to learn tasks from a single demonstration modality. This restricts the approach to be scaled to learn and execute a series of tasks in a real-life environment. In this paper, we propose a multi-modal learning approach using DMP+ with linear decay integrated in a dialogue system with speech and ontology for the robot to learn seamlessly through natural interaction modalities (like an apprentice) while learning or re-learning is done on the fly to allow partial updates to a learned task to reduce potential user fatigue and operational downtime in teaching. The performance of new DMP+ with linear decay system is statistically benchmarked against state-of-the-art DMP implementations. A gluing demonstration is also conducted to show how the system provides seamless learning of multiple tasks in a flexible manufacturing set-up.

This paper presents a technique to model statically indeterminate forces based on stiffness matrices for multi-limbed climbing robots. Current wall climbing robots in literature overlook statically indeterminate forces, causing an incapability to estimate climbing failure under certain circumstances. Accounting for these forces, robot deformation can be approximated, paving the way for the proposed two-wall climbing approach. During a wall climb, two failure modes, slide and over-torque, are identified to compute feasible climbing region. A hexapod robot is used to verify the proposed technique by climbing between walls with pure friction end effectors.

This paper proposes an overall Multi-Controller Architecture (MCA) for safe and flexible navigation of autonomous navigation, under uncertainties in highway use-cases. In addition to the details given about the main modules (and their interactions) composing the proposed MCA, an important focus of the paper is made on the definition of a robust Two-Sequential Level Decision Network (TSLDN), which uses both: Extended Time-To-Collision (ETTC) metric and a new definition of a specific Predicted Inter-Distance Profile (PIDP, between vehicles during lane changes maneuvers) in order to estimate the maneuvers risks. The TSLDN is utilized for: the driving situation assessment, decision-making and for safety retrospection over the current maneuver risk. It allows us to have the best decision to achieve the vehicle navigation task while maximizing its safety. Several simulation results show the good performance of the overall proposed control architecture, mainly in terms of efficiency to handle probabilistic decision-making even for very risky scenarios.

In this paper, a novel 3D coverage path planning method, which is efficient and practical for inspection of high-rise structures such as buildings or towers, using an unmanned aerial vehicle (UAV) is presented. Our approach basically focuses on developing a model-based path planner for structural inspection with a prior map, which is opposite to a non-model based exploration. The proposed method uses a volumetric map which is made before the path planning. With the map, the whole structure is divided into several layers for efficient path planning. Firstly, in each layer, a set of the normal vectors of the center point of every voxel is calculated, and then the opposing vectors become viewpoints. Due to too many viewpoints and an overlapped inspection surface, we down-sample them with a voxel grid filter. Then, the shortest tour connecting the reduced viewpoints must be computed with the Traveling Salesman Problem (TSP) solver. Lastly, all the paths in each layer are combined to form the complete path. The results are verified using simulations with a rotary wing UAV and compared with other state-of-the-art algorithm. It is proven that our method performs much better for structural inspection with respect to computation time as well as the coverage completeness.

This work presents a model-based approach for creating robust control policies for rolling locomotion with a spherical tensegrity topology. Utilizing the structured dynamics of Class-1 tensegrity systems, we turn to model predictive control (MPC) to generate optimal multi-cable actuation trajectories for dynamic rolling. Although the resulting multi-cable state-action trajectories successfully outperform the benchmark single-cable policy performance in speed, computational constraints prevent MPC from being applied in real-time. To address this, we demonstrate that a contextual policy trained using supervised deep learning on the generated optimal MPC trajectories can be used as an end-to-end feedback policy for real-time directed rolling locomotion.

This work presents a strategy for coordinated multi-agent weeding under conditions of partial environmental information. The goal of this work is to demonstrate the feasibility of coordination strategies for improving the weeding performance of autonomous agricultural robots. We show that, given a sufficient number of agents, the algorithm can successfully weed fields with various initial seed bank densities, even when multiple days are allowed to elapse before weeding commences. Furthermore, the use of coordination between agents is demonstrated to strongly improve system performance as the number of agents increases, enabling the system to eliminate all the weeds in the field, as in the case of full environmental information, when the planner without coordination failed to do so. As a domain to test our algorithms, we have developed an open source simulation environment, Weed World, which allows real-time visualization of coordinated weeding policies, and includes realistic weed generation. In this work, experiments are conducted to determine the required number of agents and their required transit speed, for given initial seed bank densities and varying allowed days before the start of the weeding process.

Simulation is an appealing option for validating the safety of autonomous vehicles. Generative Adversarial Imitation Learning (GAIL) has recently been shown to learn representative human driver models. These human driver models were learned through training in single-agent environments, but they have difficulty in generalizing to multi-agent driving scenarios. We argue these difficulties arise because observations at training and test time are sampled from different distributions. This difference makes such models unsuitable for the simulation of driving scenes, where multiple agents must interact realistically over long time horizons. We extend GAIL to address these shortcomings through a parameter-sharing approach grounded in curriculum learning. Compared with single-agent GAIL policies, policies generated by our PS-GAIL method prove superior at interacting stably in a multi-agent setting and capturing the emergent behavior of human drivers.

We present Move Base Flex (MBF), a highly flexible, modular, map-independent, open-source navigation framework for use in ROS. MBF provides modular actions for executing plugins for path planning, motion control, and recovery. These actions define interfaces for external executives to allow highly flexible navigation strategies, which can be intertwined with other robot tasks. MBF has been successfully deployed in a professional setting at customer facilities to control robots in highly dynamic environments. We compare MBF with the well-known move_base and present the architecture as well as different deployment approaches, including how MBF can be used with different executives to perform complex navigation tasks interleaved with other robot operations.

Intervention Autonomous Underwater Vehicle or I-AUV has recently started to grab researchers attention in the last 20 years. Only three I-AUVs have demonstrated autonomous manipulation skills: ALIVE, SAUVIM and GIRONA 500. While prior systems rely on variations of the task-priority redundancy control framework, our recent research showed preliminary results using motion planning for floating-based intervention in the presence of obstacles. With the increasing need for autonomously performing more complex manipulation tasks, two main challenges need to be addressed: the high-dimensionality of the system, and the motion coordination between the mobile base and the working arm. The latter challenge is of high importance if accurate execution is required, especially considering the floating nature of the AUV and the control challenges that come with it. Our approach relies on exploiting the loose coupling between the AUV and the arm. In particular we present an approach based on MR-MHA * (Multi-Representation, Multi-Heuristic A*), and we show how it can generate efficient trajectories by exploiting decoupling. We show for the first time the use of a search-based planner on a high-dimensional underwater manipulator. In addition, we support our claims with experimental analysis of the generated trajectories with respect to various metrics in different environments. Furthermore, we demonstrate the ability of our approach to conduct a full intervention mission in a realistic simulated underwater intervention environment.

This paper develops and compares two motion planning algorithms for a tethered UAV with and without the possibility of the tether contacting the confined and cluttered environment. Tethered aerial vehicles have been studied due to their advantages such as power duration, stability, and safety. However, the disadvantages brought in by the extra tether have not been well investigated by the robotic locomotion community, especially when the tethered agent is locomoting in a non-free space occupied with obstacles. In this work, we propose two motion planning frameworks that (1) reduce the reachable configuration space by taking into account the tether and (2) deliberately plan (and relax) the contact point(s) of the tether with the environment and enable an equivalent reachable configuration space as the non-tethered counterpart would have. Both methods are tested on a physical robot, Fotokite Pro. With our approaches, tethered aerial vehicles could find their applications in confined and cluttered environments with obstacles as opposed to ideal free space, while still maintaining the advantages from the usage of a tether. The motion planning strategies are particularly suitable for marsupial heterogeneous robotic teams, such as visual servoing/assisting for another mobile, tele-operated primary robot.

A motion planner is developed for guiding a small aerobatic fixed-wing unmanned aerial vehicle to a desired goal region in a highly constrained, three-dimensional, known environment with static obstacles. The planner is based on the Rapidly-Exploring Random Trees (RRT) algorithm, and pieces together feasible trajectories from a library of motion primitives. Among other more conventional motion primitives, the library includes three extreme maneuvers: a cruise-to-hover transition, a hover-to-cruise transition, and an aggressive turn-around. The algorithm is efficient; it can be run in real-time to rapidly generate a plan starting from the aircraft's configuration at run-time. The motion planner is closely coupled to a feedback controller. Simulations using an aircraft dynamics model demonstrate the effectiveness of the system to guide and control the aircraft to a desired goal region. Preliminary flight test results are also presented.

This paper is concerned with automating fleets of autonomous robots. This involves solving a multitude of problems, including goal assignment, motion planning, and coordination, while maximizing some performance criterion. While methods for solving these sub-problems have been studied, they address only a facet of the overall problem, and make strong assumptions on the use-case, on the environment, or on the robots in the fleet. In this paper, we formulate the overall fleet management problem in terms of Optimal Control. We describe a scheme for solving this problem in the particular case of fleets of non-holonomic robots navigating in an environment with obstacles. The method is based on a two-phase approach, whereby the first phase solves for fleet-wide boolean decision variables via Mixed Integer Quadratic Programming, and the second phase solves for real-valued variables to obtain an optimized set of trajectories for the fleet. Examples showcasing the features of the method are illustrated, and the method is validated experimentally.

Robots that navigate among pedestrians use collision avoidance algorithms to enable safe and efficient operation. Recent works present deep reinforcement learning as a framework to model the complex interactions and cooperation. However, they are implemented using key assumptions about other agents' behavior that deviate from reality as the number of agents in the environment increases. This work extends our previous approach to develop an algorithm that learns collision avoidance among a variety of types of dynamic agents without assuming they follow any particular behavior rules. This work also introduces a strategy using LSTM that enables the algorithm to use observations of an arbitrary number of other agents, instead of previous methods that have a fixed observation size. The proposed algorithm outperforms our previous approach in simulation as the number of agents increases, and the algorithm is demonstrated on a fully autonomous robotic vehicle traveling at human walking speed.

Task level programming based on skills has often been proposed as a mean to decrease programming complexity of industrial robots. Several models are based on encapsulating complex motions into self-contained primitive blocks. A semantic skill is then defined as a deterministic sequence of these primitives. A major limitation is that existing frameworks do not support the coordination of concurrent motion primitives with possible interference. This decreases their reusability and scalability in unstructured environments where a dynamic and reactive adaptation of motions is often required. This paper presents a novel framework that generates adaptive behaviors by modeling skills as concurrent motion primitives activated dynamically when conditions trigger. The approach exploits the additive property of motion generators to superpose multiple contributions. We demonstrate the applicability on a real assembly use-case and discuss the gained benefits.

This paper presents a novel sliding mode controller for trajectory tracking of the piezo-driven stage. The tracking performance of piezoelectric actuator is mainly affected by the hysteresis nonlinearity. Sliding mode control is a possible solution to achieve better tracking performance. However, conventional sliding mode control generates discontinuous control signal which results in chattering. Hence, the hysteresis nonlinearity is first compensated with a hysteresis model, and an uncertainty and disturbance estimator is designed and included to devise a smooth control action. The stability of the proposed method is demonstrated via Lyapunov analysis. Both simulation and experiment are also conducted to verify the effectiveness of the proposed approach. The results are compared with a conventional sliding mode controller and a proportional-integral control with notch filter (PIC-NF).

Sensor fusion is a fundamental process in robotic systems as it extends the perceptual range and increases robustness in real-world operations. Current multi-sensor deep learning based semantic segmentation approaches do not provide robustness to under-performing classes in one modality, or require a specific architecture with access to the full aligned multi-sensor training data. In this work, we analyze statistical fusion approaches for semantic segmentation that overcome these drawbacks while keeping a competitive performance. The studied approaches are modular by construction, allowing to have different training sets per modality and only a much smaller subset is needed to calibrate the statistical models. We evaluate a range of statistical fusion approaches and report their performance against state-of-the-art baselines on both realworld and simulated data. In our experiments, the approach improves performance in IoU over the best single modality segmentation results by up to 5%. We make all implementations and configurations publicly available.