Consider the following world: teams of autonomous robots have become routinely used in adversarial situations, such as searching a building for a bomb. A clever adversary would probably try to infiltrate the robot team to sabotage its efforts, either with their own masked robot or by taking over one of the team robots. Could such a traitor be detected?
For this project, algorithms were developed to visually analyze robot behaviors and detect traitors within multi-robot teams. The problem was simplified to consider only on robot behaviors that are not influenced by their environment, mainly to avoid the problem of trying to infer what other robots' sensor input is. This kept the problem tractable for an initial investigation. A metric was created to determine how similar an observed behavior is to an expected behavior, and simulations and real-world experiments were run. The results were good.
- Nathaniel Bird and Nikolaos Pananikolopoulos, "Recognition of Traitors in Distributed Robotic Teams," Proceedings of the IEEE International Conference on Robotics and Automation, May 2011.