Goal-oriented action planning in partially observable stochastic domains

Partially Observable Markov Decision Processes (POMDPs) provide a rich framework for sequential decision-making under uncertainty in stochastic domains. The paper presented a probabilistic conditional planning problem for Goal-Oriented Action Planning based on POMDP (called p-GOAP). We are interested in finding a plan such that the plan has maximal the goal satisfaction subject to the cost not exceeding the threshold in p-GOAP. During computing maximum goal satisfaction, we discuss a speed-up technique that alleviates the computational complexity by separating the algorithm into two phases: a greedy algorithm and a recursive process. Finally p-GOAP is proposed to cognitive reappraisal for deliberate emotion.