Probabilistic Physical Search on General Graphs: Approximations and Heuristics

An autonomous intelligent agent often needs to explore its environment and choose among different available alternatives. In many physical environments the exploration is costly, and the agent also faces uncertainty regarding the price of the possible alternatives. For example, consider a traveling purchaser seeking to obtain an item [7]. While there may be prior knowledge regarding candidate stores (e.g., based on search history), the actual price at any given site may only be determined upon reaching the site. In another domain, consider a Rover robot seeking to mine a certain mineral on the face of Mars [8, 9]. While there may be prior knowledge regarding candidate mining sites (e.g., based on satellite images) [3, 4], the actual cost associated with the mining at any given location, e.g., in terms of battery consumption, may depend on the exact conditions at each site (e.g., soil type, terrain, etc.), and hence are fully known only upon reaching the site.