OCP: An Online Contingent Planning Method for Robot Tasks with Incomplete Knowledge

Autonomous robots face challenges posed by uncertain states when operating in environments with incomplete knowledge. Such uncertainty complicates accurate state representation and efficient task planning. In the field of robotics, both contingent planning and probabilistic planning have proven effective in generating robust plans under such conditions. However, as the scale of the problem escalates, these approaches inevitably yield numerous redundant plans, thereby resulting in low planning efficiency. To counter these issues, we propose an Online Contingent Planning (OCP) method that augments the Planning Domain Definition Language with two new tokens and integrates an innovative online contingent planner. By incorporating these tokens, OCP is able to denote multiple uncertain states within a single state and defer the determination of the state until the execution phase. The dynamic feature of OCP to update the plans during execution eliminates a large portion of redundant plans and planning time. Experimental evaluations demonstrate that OCP substantially surpasses the planning efficiency of several state-of-the-art planners with lower plan size, marking a significant advancement in autonomous robot task planning.