Countering attacks and failures through resilient coordination
Most of the research on robot resiliency focuses on countering deceptive attacks that mislead the robot team. Instead, we have so far focused against denial-of-service (DoS) attacks and failures that can make robots fail or compromise their sensors. We aimed at guaranteeing team performance even if some robots in the team get DoS attacks. To this end, we formulated game-theoretic problems between the robots and the adversary, and designed the first provably near-optimal approximation algorithms for robust combinatorial optimization in various settings including centralized, decentralized communication and short-horizon, long-horizon planning. These algorithms enabled DoS-resilient multi-robot planning in data collection scenarios such as target tracking and environmental exploration. Apart from sensor attacks, the communications among robots can be easily jammed and disrupted by the adversary. Thus, we have recently investigated near-optimal resilient algorithms to protect the team performance from both sensor and communication attacks/failures.
Besides the aforementioned resilient algorithms that can withstand attacks/failures, I am also interested in how robots should react to and recover from attacks/failures. To this end, we developed a resilient coordination framework that enables robots to adapt and recover by reconfiguring team resources to compensate for the performance loss induced by robot failures.
RA-L+ICRA’19: Robust submodular maximization against robot/sensor attacks/failures;
ICRA’20, T-RO: Decentralized (clique-based) robust submodular maximization against robot/sensor attacks/failures (left video);
RA-L’21: Distributed (consensus-based) robust submodular maximization against robot/sensor attacks/failures;
RSS’20, full version: Robust team orienteering over a longer planning time against robot/sensor attacks/failures;
ACC’22 invited paper, full version: Robust monotone maximization against robot/sensor and communication attacks/failures;
IROS’20: Resilient coordination to adapt to and recover from robot/sensor failures (right video).