Planning with Participation Constraints

We pose and study the problem of planning in Markov decision processes (MDPs), subject to participation constraints as studied mechanism design. In this problem, a planner must work with self-interested agent on given MDP. Each action MDP provides an immediate reward (possibly different) agent. The has no control choosing actions, but option end entire process at any time. goal is find policy that maximizes her cumulative reward, taking into consideration agent's ability terminate. give fully polynomial-time approximation scheme for problem. En route, we present algorithms computing (exact) optimal policies important special cases including when time horizon constant, or exhibits "definitive decisions" property. illustrate our two different game-theoretic applications: assigning rides ride-sharing designing screening policies. Our results imply efficient (approximately) both applications.