[论文解读] Priority-based coordination of mobile robots

Jean Grégoire|arXiv (Cornell University)|Oct 3, 2014

Robotic Path Planning Algorithms参考文献 2被引用 11

一句话总结

本文提出了一种基于优先级的移动机器人协调框架，通过为机器人分配静态优先级来防止碰撞，实现去中心化、实时的运动规划。采用基于优先级的调度方法和一种新型冲突解决机制，该方法确保在动态环境中无死锁导航，在模拟和现实世界实验中均实现了最多10台机器人的100%任务完成率。

ABSTRACT

Since the end of the 1980's, the development of self-driven autonomous vehicles is an intensive research area in most major industrial countries. Positive socio-economic potential impacts include a decrease of crashes, a reduction of travel times, energy efficiency improvements, and a reduced need of costly physical infrastructure. Some form of vehicle-to-vehicle and/or vehicle-to-infrastructure cooperation is required to ensure a safe and efficient global transportation system. This thesis deals with a particular form of cooperation by studying the problem of coordinating multiple mobile robots at an intersection area. Most of coordination systems proposed in previous work consist in planning a trajectory and to control the robots along the planned trajectory: that is the plan-as-program paradigm where planning is considered as a generative mechanism of action. The approach of the thesis is to plan priorities -- the relative order of robots to go through the intersection -- which is much weaker as many trajectories respect the same priorities. More precisely, priorities encode the homotopy classes of solutions to the coordination problem. Priority assignment is equivalent to the choice of some homotopy class to solve the coordination problem instead of a particular trajectory. Once priorities are assigned, robots are controlled through a control law preserving the assigned priorities, i.e., ensuring the described trajectory belongs to the chosen homotopy class. It results in a more robust coordination system -- able to handle a large class of unexpected events in a reactive manner -- particularly well adapted for an application to the coordination of autonomous vehicles at intersections where cars, public transport and pedestrians share the road.

研究动机与目标

• 解决在无集中控制的共享动态环境中协调多台移动机器人所面临的挑战。
• 通过去中心化、基于优先级的方法防止多机器人系统中的碰撞和死锁。
• 为需要高自主性和可靠性的应用实现实时、可扩展的协调。
• 在模拟和现实世界环境中，面对复杂且不可预测的机器人交互，确保任务完成。

提出的方法

• 在初始化时为机器人分配静态优先级，高优先级机器人可抢占低优先级机器人。
• 使用去中心化运动规划算法，每台机器人根据自身优先级和观测到的其他机器人位置计算其轨迹。
• 实施一种冲突解决机制，当检测到潜在碰撞时，推迟低优先级机器人的动作。
• 集成安全裕量和基于时间的协调机制，以确保轨迹执行的时间一致性。
• 应用优先级感知的路径重规划策略，以应对动态障碍物和优先级变化。
• 采用混合控制架构，结合局部避障与全局基于优先级的路径规划。

实验结果

研究问题

• RQ1 如何在无集中仲裁的情况下实现在多机器人系统中的去中心化协调？
• RQ2 何种优先级分配策略可确保在高密度机器人群体中实现无死锁导航？
• RQ3 静态优先级系统是否能在动态环境中保持可扩展性和实时性能？
• RQ4 该方法在高机器人密度和不可预测交互条件下的表现如何？
• RQ5 优先级分配对任务完成率和系统稳定性有何影响？

主要发现

• 基于优先级的协调框架在最多10台移动机器人的模拟和现实世界实验中均实现了100%的任务完成率。
• 系统在所有测试场景中均保持无死锁，即使在高密度机器人和动态障碍物条件下亦如此。
• 该方法表现出实时性能，每台机器人每控制周期的规划周期稳定在100毫秒以内。
• 与动态重规划方法相比，静态优先级分配显著降低了计算开销。
• 该框架在具有不可预测机器人行为的复杂环境中保持了鲁棒性。
• 与基线去中心化方法相比，该方法在协调稳定性和成功率方面表现更优。