Cloud-based Robot Path Planning in Dynamic Environments

摘要

Most of the current path planning applications only focus on a single agent, without considering the surrounding robots or dynamic obstacles. In safety-critical environments, such as the crowded playground, the robot may have increased difficulty reaching the destination with crowds moving around, sometimes even cause damages. With the development of cloud computing and swarm intelligence, the concept of “cloud robotic” has been followed by more scholars. With collaboration and information sharing between robots, the cluster has better problem-solving skills than the individual. In this paper, we propose a cloud-based multi-agent navigation algorithm in dynamic environments. We propose the concept of pheromones of dynamic obstacles that represent congestion to enable clusters to plan collaboratively. Widespread static sensors are introduced to jointly estimate congestion in the environment with robots. When an agent needs route planning, the cloud provides safe and fast route with intermediate points. The robot uses a simple yet effective local planner based on artificial potential field (APF) to trace the trajectory. We demonstrate this approach's effectiveness compared to traditional A* global planner and APF local planner with traditional repulsion function. Experiments show that our cloud-based method reduces the number of collisions by 92.6%, with only 35% increase in path length. All code for reproducing the experiments is at https://github.com/Asber777/CDPP.