Hiding Leader's Identity in Leader-Follower Navigation through Multi-Agent Reinforcement Learning

Leader-follower navigation is a popular class of multi-robot algorithms where\na leader robot leads the follower robots in a team. The leader has specialized\ncapabilities or mission-critical information (e.g. goal location) that the\nfollowers lack, and this makes the leader crucial for the mission's success.\nHowever, this also makes the leader a vulnerability - an external adversary who\nwishes to sabotage the robot team's mission can simply harm the leader and the\nwhole robot team's mission would be compromised. Since robot motion generated\nby traditional leader-follower navigation algorithms can reveal the identity of\nthe leader, we propose a defense mechanism of hiding the leader's identity by\nensuring the leader moves in a way that behaviorally camouflages it with the\nfollowers, making it difficult for an adversary to identify the leader. To\nachieve this, we combine Multi-Agent Reinforcement Learning, Graph Neural\nNetworks and adversarial training. Our approach enables the multi-robot team to\noptimize the primary task performance with leader motion similar to follower\nmotion, behaviorally camouflaging it with the followers. Our algorithm\noutperforms existing work that tries to hide the leader's identity in a\nmulti-robot team by tuning traditional leader-follower control parameters with\nClassical Genetic Algorithms. We also evaluated human performance in inferring\nthe leader's identity and found that humans had lower accuracy when the robot\nteam used our proposed navigation algorithm.\n