强化学习在无人车领域的应用与展望

无人车(UGV)可替代人类自主地执行民用和军事任务,对未来智能交通及陆军装备发展有重要战略意义。随着人工智能技术的日益成熟,采用强化学习技术成为了无人车智能决策领域最受关注的发展趋势之一。本文首先简要概述了强化学习的发展历程、基础原理和核心算法;随后,分析总结了强化学习在无人车智能决策中的研究进展,包括障碍物规避、变道与超车、车道保持和道路交叉口通行四种典型场景;最后,针对基于强化学习的智能决策面临的问题和挑战,探讨并展望了未来的研究工作与潜在的研究方向。     

基于梯形棋盘格的摄像机和激光雷达标定方法

针对无人车（UGV）自主跟随目标车辆检测过程中需要对激光雷达（LiDAR）数据和摄像机图像进行信息融合的问题,提出了一种基于梯形棋盘格标定板对激光雷达和摄像机进行联合标定的方法。首先,利用激光雷达在梯形标定板上的扫描信息,获取激光雷达安装的俯仰角和安装高度;然后,通过梯形标定板上的黑白棋盘格标定出摄像机相对于车体的外参数;其次,结合激光雷达数据点与图像像素坐标之间的对应关系,对两个传感器进行联合标定;最后,综合激光雷达和摄像机的标定结果,对激光雷达数据和摄像机图像进行了像素级的数据融合。该方法只要让梯形标定板放置在车体前方,采集一次图像和激光雷达数据就可以满足整个标定过程,实现两种类型传感器的标定。实验结果表明,该标定方法的平均位置偏差为3.5691 pixel,折算精度为13 μm,标定精度高。同时从激光雷达数据和视觉图像融合的效果来看,所提方法有效地完成激光雷达与摄像机的空间对准,融合效果好,对运动中的物体体现出了强鲁棒性。     

IoT-Enabled Autonomous System Collaboration for Disaster-Area Management

Timely investigating post-disaster situations to locate survivors and secure hazardous sources is critical, but also very challenging and risky. Despite first responders putting their lives at risk in saving others, human-physical limits cause delays in response time, resulting in fatality and property damage. In this paper, we proposed and implemented a framework intended for creating collaboration between heterogeneous unmanned vehicles and first responders to make search and rescue operations safer and faster. The framework consists of unmanned aerial vehicles (UAVs), unmanned ground vehicles (UGVs), a cloud-based remote control station (RCS). A light-weight message queuing telemetry transport (MQTT) based communication is adopted for facilitating collaboration between autonomous systems. To effectively work under unfavorable disaster conditions, antenna tracker is developed as a tool to extend network coverage to distant areas, and mobile charging points for the UAVs are also implemented. The proposed framework’s performance is evaluated in terms of end-to-end delay and analyzed using architectural analysis and design language (AADL). Experimental measurements and simulation results show that the adopted communication protocol performs more efficiently than other conventional communication protocols, and the implemented UAV control mechanisms are functioning properly. Several scenarios are implemented to validate the overall effectiveness of the proposed framework and demonstrate possible use cases.      

平行机器人与平行无人系统:框架、结构、过程、平台及其应用

本文将基于ACP（Artificial societies,computational experiments,parallel execution）的平行系统思想与机器人领域相结合,形成一种软硬件相结合的框架,为无人机、无人车、无人船在复杂环境中实验、学习与实际工作提供便捷、安全的平台,即平行无人系统.本文从平行机器人的基本概念出发,提出平行无人系统的基本框架,并介绍了各模块的基本功能与实现方法,探讨了其中的关键技术.然后本文围绕无人机、无人车、无人船三个方面展望了无人平行系统在实际中的应用和所面临的挑战,提出了平行无人系统的未来发展方向.     

Vehicle Teleoperation Interfaces

Despite advances in autonomy, there will always be a need for human involvement in vehicle teleoperation. In particular, tasks such as exploration, reconnaissance and surveillance will continue to require human supervision, if not guidance and direct control. Thus, it is critical that the operator interface be as efficient and as capable as possible. In this paper, we provide an overview of vehicle teleoperation and present a summary of interfaces currently in use.     

室外崎岖地形下基于视差图的无人自主车障碍物识别

对于地面无人车和室外非结构化环境, 本文介绍了我们开发的基于立体视觉的障碍物快速识别系统. 为了使无人地面车适应于较复杂的地形, 根据V视差图, 我们提出了一种新的地面主视差图的估计方法. 通过地面主视差图和局部的三维重建, 本文给出了一种由粗到精的障碍物识别与定位方法. 在我们的无人地面车平台上, 我们对这一障碍物自动识别系统进行了相应的实际试验. 其试验结果验证了该系统的有效性.     

UGV协同系统研究进展*

无人地面车辆（UGV）在工业自动化、星球探索、灾后救援、智能交通以及军事作战等多任务领域都具有广阔的应用前景。UGV协同系统通过嵌入组织架构、合作策略、交互机制等协同内容,可以达到拓展环境感知范围、提高复杂环境理解适应能力和增强复杂任务工作效能的目的,受到了国内外广泛的关注。从UGV单体/群体体系结构、多重任务协同分配方法以及协同定位、编队、覆盖/探索等几个方面对目前国内外UGV协同工作关键技术进行了总结,给出了一个UGV协同系统的应用实例,并指出了系统发展趋势。     

强化学习的地-空异构多智能体协作覆盖研究

以无人机（unmanned aerial vehicle, UAV）和无人车（unmanned ground vehicle, UGV）的异构协作任务为背景,通过UAV和UGV的异构特性互补,为了扩展和改进异构多智能体的动态覆盖问题,提出了一种地-空异构多智能体协作覆盖模型。在覆盖过程中,UAV可以利用速度与观测范围的优势对UGV的行动进行指导;同时考虑智能体的局部观测性与不确定性,以分布式局部可观测马尔可夫（decentralized partially observable Markov decision processes,DEC-POMDPs）为模型搭建覆盖场景,并利用多智能体强化学习算法完成对环境的覆盖。仿真实验表明,UAV与 UGV间的协作加快了团队对环境的覆盖速度,同时强化学习算法也提高了覆盖模型的有效性。     

基于Vericut的数控程序仿真技术研究

由于宏程序中含有逻辑运算和数学运算,变量的变化很难凭肉眼观察出来,无法便捷地验证程序的正确性。针对验证宏程序比较困难的现状,本文基于Vericut仿真软件,介绍了在宏程序调试过程中,通过观察切削图形和变量变化,快速找出错误原因的方法。另一方面,大部分数控编程软件具有刀路仿真功能,但是通常仿真功能比较简单,如果用户想对加工后的零件模型进行进一步分析,CAM软件就显得无能为力了。本文同时介绍了通过UGV接口文件将UG/CAM中的数据传递到Vericut中进行仿真验证的途径。     

Improving Endurance and Range of a UGV with Gimballed Landing Platform for Launching Small Unmanned Helicopters

Design specifications for a high-endurance and range unmanned ground vehicle (UGV) with a gimballed landing platform on top of it for takeoff/landing, transporting to the target area and recharging of small/miniature unmanned helicopters are presented and justified. Specification constraints include UGV strict payload limitations, limited free space affecting power supply availability that impacts on-board available energy, limited endurance and operational range, as well as limitations and restrictions related to electric and non-electric small unmanned vertical takeoff and landing (VTOL) vehicles, similar to those of UGV with the most important being limited flying time. Focusing on the All Terrain Robot Vehicle (ATRV-Jr) UGV and a helicopter of the size of the Maxi Joker 2 as a testbed, a detailed analysis of component power consumption reveals reasons for reduced runtime and operational range. After a comparative study of state of the art power supply and battery technologies, a hybrid battery configuration is proposed that improves more than 500% the manufacturer-specified ATRV-Jr endurance (or 1,000% the currently used custom-made ATRV-Jr endurance) by considering: (1) optimum design with weight, volume, runtime and rechargeability being major restrictions and concerns, and (2) use of lower power sensors and processors without affecting UGV functionality and operability. A sun-tracking solar array that collects and stores energy is integrated with the UGV gimballed landing platform. Simulations demonstrate the validity of the design. Although the testbed is specific, the design itself is generic enough and suitable for other UGV/VTOL vehicles. This work was supported by the US Army Research Office, Grant Number W91-11NF-06-1-0069 and SPAWAR, Grant Number N00039-06-C-0062.