An Issue in Goal Addition in Continuous Robotic Plan Execution

Robotic plan execution has traditionally assumed that goals are articulated prior to mission execution. As robots have become persistent and increasingly moved into real-world environments, this assumption is not necessarily true; for instance a user can decide to give a new objective to the robot for inclusion in the plan being formulated, add newer goals, or modify others queued for execution. In most systems this leads, at best, to a suboptimal final plan or possibly to the exclusion of objectives, either of which could have been avoided, should the robot have executed its initial plan differently. We first articulate and then demonstrate a preliminary approach to this problem motivated by a marine robotics domain. We do so with an execution policy that is sufficient to disambiguate actions for execution within a flexible temporal continuous plan execution system. The resulting algorithmic complexity is linear in the number of actions and causal links of an existing partial plan.