Collisions as Information Sources in Robot Swarms

Collision avoidance is crucial for making multi-robot systems operate in a safe manner–to keep the robots safe, collisions should be avoided at all costs. In a number of robotic applications, collisions have the potential to be catastrophic, e.g., for fleets of unmanned aerial vehicles or platoons of self-driving trucks. However, as the density of robots operating in a given space increases, robots spend more time avoiding each other as opposed to progressing the overall team mission.

In this project, we sidestep this issue completely and embrace collisions as a feature of the multi-robot system. In fact, as momentum is velocity times mass, collisions become less of an issue as the robots get  smaller and slower. Furthermore, there are plenty of examples in nature where collisions among individuals are not only tolerated, but are also used as a sensory mechanism to obtain information about the swarm. For example, ants frequently engage in head-to-head contact to regulate traffic flow and ensure efficient transport of resources in tunnels and narrow trails.

Inspired by such phenomena, we investigate whether or not collisions can be used as sufficiently rich sources of information to enable a team of robots to localize themselves (in particular environments). Given a domain which is divided into a fixed number of cells, we develop a probabilistic localization technique which allows individual robots to estimate which cell they are currently present in using binary collision information. To this end, we develop analytical collision models to describe the nature and frequency of collisions experienced by a robot in the swarm as a function of the robot density. Collision measurements are then incorporated via a hidden Markov model framework to allow each robot to localize itself. The following video describes the localization algorithm implemented on a team of robots on the Robotarium:


Investigators:

  • Siddharth Mayya
  • Pietro Pierpaoli
  • Girish Nair
  • Magnus Egerstedt

Related Publications:

Siddharth Mayya, Pietro Pierpaoli, Girish Nair, and Magnus Egerstedt, “Localization in densely packed swarms using interrobot collisions as a sensing modality”, IEEE Transactions on Robotics, vol. 35, no. 1, pp. 21–34, Feb 2019.

Siddharth Mayya, Pietro Pierpaoli, Girish Nair, and Magnus Egerstedt, “Collisions as Information Sources in Densely-Packed Robot Swarms Under Mean-Field Approximations”, Proceedings of Robotics: Science and Systems, 2017.

 

 

Comments are closed.