Decentralized risk sharing in teams of unmanned vehicles

We investigate the problem of planning for a team of unmanned agents in the presence of constraints. The agents are allowed some probability of violating the constraints, a quantity defined as “risk”. The total risk allowed for the team must be allocated among the constituent agents, and in this work we propose an auctioning mechanism through which agents can bid for risk. Whether a particular bid results in a team performance improvement or not is decided by the agents through consensus. Simple examples suggest a significant improvement in performance, as measured by a reward model, when compared to fixed risk allocations. Monte Carlo simulation results are presented showing an improvement in team performance particularly in environments with tight constraints.