Decentralized path planning for multi-agent teams with complex constraints

Abstract

This paper presents a novel approach to address the challenge of planning paths for multi-agent systems subject to complex constraints. The technique, called the Decentralized Multi-Agent Rapidly-exploring Random Tree (DMA-RRT) algorithm, extends the Closed-loop RRT (CL-RRT) algorithm to handle multiple agents while retaining its ability to plan quickly. A core component of the DMA-RRT algorithm is a merit-based token passing coordination strategy that makes use of the tree of feasible trajectories grown in the CL-RRT algorithm to dynamically update the order in which agents replan. The reordering is based on a measure of each agent's incentive to change the plan and allows agents with a greater potential improvement to replan sooner, which is demonstrated to improve the team's overall performance compared to a traditional, scripted replan order. The main contribution of the work is a version of the algorithm, called Cooperative DMA-RRT, which introduces a cooperation strategy that allows an agent to modify its teammates' plans in order to select paths that reduce their combined cost. This modification further improves team performance and avoids certain common deadlock scenarios. The paths generated by both algorithms are proven to satisfy inter-agent constraints, such as collision avoidance, and numerous simulation and experimental results are presented to demonstrate their performance.

Subdimensional expansion for multirobot path planning

“…Velocity planners fix the paths that will be followed by each robot, then find a velocity schedule along those paths that avoids collisions [5,29,30,31,32,33,34]. Priority planners assign a priority to each robot, then plan for individual robots in decreasing order of priority, treating higher priority robots as moving obstacles [4,35,36,37,38,39]. The choice of priority ordering is critical, leading to a number of heuristics for choosing priority orders that are likely to lead to a solution [38,40,41].…”

Group abstraction for assisted navigation of social activities in intelligent environments

“…In classical robotics, robots are typically under full control of the motion planner, so it is assumed that they will closely follow the plan or actions deemed most likely to achieve the desired goal [6,14,10,9,39]. Even in the case of disruptions or problems, robots can be forced to take a particular path, or wait indefinitely until it is "safe" to proceed [10,9,39].…”

“…Even compared to robotic systems that must consider disruptions [14,39], the variability of the environment and disruptions envisaged in the ACANTO context may be significantly higher.…”

Dynamic Path Planning for Mobile Robots with Cellular Learning Automata

“…Designing an optimal path is a difficult task [1,2]. There are many aspects that must be considered, such as the model of the environment, existing constraints, kinematics and the dynamic properties of the robot itself.…”