A Potential Field Approach To Multiple Robot Formation Control

A special case of cooperative control for mobile robots is considered - formation control. A potential field based algorithm is developed in which geometric, communication and information centric influences are considered and allowed to deform the formation. Control is accomplished in a leader-follower(s) method. One leader robot defines the overall trajectory and follower nodes individually and autonomously maintain the formation while simultaneously moving toward a goal position. The particular nodes considered are the ARRIbots developed at the Distributed and Intelligence and Autonomy Lab (DIAL) in the Automation and Robotic and Research Institute (ARRI), which are non-holonomic differential-drive wheeled robots. In order to facilitate the above, a kinematic model for the ARRIbot mobile robot that accounts for its non-holonomic constraint and particular inputs is presented. Based on this model, a trajectory tracking controller (LQR based) is detailed assuming a given reference trajectory. Finally additional artificial forces are added to account for additional constraints on the optimal node paths and positions. Obstacle avoidance is added to the formation by repulsive forces and an artificial communications force is added in order to optimize the wireless communications channel between nodes. These results are validated using computer simulations and experiments with the ARRIbots on our mobile robot platform. It was found that the potential field algorithm was successful in maintaining the required node formation. The position error for follower nodes was found to decay as required. In addition, a simulation of mine-field detection scenario was shown to successfully combine the three formation influences. The LQR trajectory tracker was found to satisfy the requirements of the formation control algorithm. The state estimate error was low for both straight line and turning motions while the tracking error was low for straight line and high for turning motions. This was attributed to the short duration of the latter and amplification of calculation and processing delay errors. The accuracy of the state estimate errors was shown to be useful in reducing the tracking error over multiple trajectories.