Risk-aware Planning in Hybrid Domains: An Application to Autonomous Planetary Rovers

Abstract

Expanding robotic space exploration beyond the immediate vicinity of Earth's orbit can only be achieved by increasingly autonomous agents, given the sometimes insurmountable challenges of teleoperation over great distances. Among the numerous requirements that a fully autonomous robotic space explorer must meet, here we focus on three key mission-enabling technologies: the ability to act under uncertainty and adapt to its environment; the ability to optimize performance while offering hard bounds on the risk of mission failure; and the ability to handle complex hybrid (discrete and continuous) mission planning problems in a provably correct and scalable fashion. In this paper, we show how CLARK, a planner capable of generating risk-bounded, dynamic temporal plans for autonomous agents operating under uncertainty, is able to efficiently generate temporal plans for a challenging planetary rover scenario featuring temporal uncertainty that could not be addressed by previous methods.