Using Player Models to Improve Robustness of HTN Plans in Multi-Agent Domains

When multiple agents act concurrently towards achieving their goals in a shared environment, interactions between their actions may arise, affecting the outcome of their plans. We propose an approach to planning in such environments, termed interference robustness optimization planning that builds upon the HTN planning paradigm and extends it with explicit consideration and optimization of plan robustness. Plan robustness is calculated from the domain model and probabilistic models of other agents in the domain. A method is presented for automatic conversion of standard HTN planning tasks into planning tasks whose output maximizes plan success probability. The method is evaluated on a test domain based on a realistic multi-agent disaster relief scenario. The empirical results indicate that the effectiveness of the method depends strongly on the predictability of other agents’ behaviour and the ratio of interaction action pairs. For any values of these control parameters, the proposed method significantly outperforms standard HTN plan-