Adaptive navigation in a heterogeneous swarm robotic system
 
Frederick Ducatelle, Gianni DiCaro, Alexander Förster and Luca M.Gambardella

Abstract

We study a situation where a swarm of wheeled robots, the foot-bots, is deployed in an indoor environment to solve a foraging problem, i.e., they need to go back and forth between a source and a target location. For the navigation between the two locations, they are assisted by a swarm of flying robots that can attach to the ceiling, the eye-bots. The eye-bots are deployed beforehand and form a grid on the ceiling between source and target. From their position on the ceiling they give directional instructions to the foot-bots on the ground. Since the topology of the terrain is different on the ceiling and on the ground, eye-bots cannot derive the instructions to give based on their own sensor feedback (e.g., distance scanner, or infrared communica- tion between eye-bots). Instead, we use an iterative solution whereby eye-bots give instructions to foot-bots and then observe the behavior and feedback of footbots to adapt the instructions they give. Through this adaptive process, the heterogeneous system of eye-bots and foot- bots is able to cooperatively learn paths through the environment. Moreover, it is capable of finding shortest paths and spreading over multiple paths in case of congestion. We describe both a swarm in- telligence inspired approach and an approach using reinforcement learning. The setup described here relates to existing work on the use of sensor networks to guide robots or persons through cluttered environments. Moreover, the proposed approach shows how stigmergic reinforcement learning can be applied in swarm robotic systems.