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

Recovery policies for safe exploration of lunar permanently shadowed regions by a solar-powered rover

“…Conditional Planning for Autonomy with Risk (CLARK) is a high-level framework for generating task plans under temporal uncertainty. CLARK solves risk-bounded partially observable Markov decision processes (POMDPs) and has been demonstrated in planetary rover exploration problems [23,24]. Internally, this framework employs the p-Sulu [25] and PARIS [26] algorithms for risk-aware path planning and scheduling, respectively.…”

Constrained Multiagent Markov Decision Processes: a Taxonomy of Problems and Algorithms

“…In such domains it is important to bound or minimize the probability that an undesirable event occurs, such as reaching a dangerous state or exceeding the capacity of a resource. Problem domains where such constraints occur include the dynamic allocation of radio spectrum while minimizing the risk of collisions (Tehrani et al, 2012), as well as planning autonomous planetary exploration (Santana et al, 2016b). In the conceptualization considered in this survey we can model a chance constraint for a single agent by defining a resource consumption function that equals 1 when the undesirable event occurs, and equals 0 otherwise.…”