Online Decentralized Receding Horizon Trajectory Optimization for\n Multi-Robot systems

Abstract

A novel decentralised trajectory generation algorithm for Multi Agent systems\nis presented. Multi-robot systems have the capacity to transform lives in a\nvariety of fields. But, trajectory generation for multi-robot systems is still\nin its nascent stage and limited to heavily controlled environments. To\novercome that, an online trajectory optimization algorithm that generates\ncollision-free trajectories for robots, when given initial state and desired\nend pose, is proposed. It utilizes a simple method for obstacle detection,\nlocal shape based maps for obstacles and communication of robots' current\nstates. Using the local maps, safe regions are formulated. Based upon the\ncommunicated data, trajectories are predicted for other robots and incorporated\nfor collision-avoidance by resizing the regions of free space that the robot\ncan be in without colliding. A trajectory is then optimized constraining the\nrobot to remain within the safe region with the trajectories represented by\npiecewise polynomials parameterized by time. The algorithm is implemented using\na receding horizon principle. The proposed algorithm is extensively tested in\nsimulations on Gazebo using ROS with fourth order differentially flat aerial\nrobots and non-holonomic second order wheeled robots in structured and\nunstructured environments.\n