MPTP: Motion-planning-aware task planning for navigation in belief space

Minimum Displacement Motion Planning for Movable Obstacles

“…Another related area is the field of task and motion planning [2,5,8,14,17,18]. Selecting the obstacles to be displace may be accomplished via task planning and the geometric constraints may be incorporated into the symbolic planner [14] or both the task and motion planning could be integrated by means of an efficient mapping between the two domains [2,18]. Manipulation among clutter or rearrangement planning in clutter [3,9,10,16] also bears resemblance to the problem proposed herein.…”

“…NAMO is proved to be NP-hard [20] and most approaches solve a subclass of problems that selects a set of obstacles to be moved and displaces them to reconfigure the environment. Another related area is the field of task and motion planning [2,5,8,14,17,18]. Selecting the obstacles to be displace may be accomplished via task planning and the geometric constraints may be incorporated into the symbolic planner [14] or both the task and motion planning could be integrated by means of an efficient mapping between the two domains [2,18].…”

Visually Grounded Task and Motion Planning for Mobile Manipulation

“…However, when there are actions that take significant time to execute (e.g., long-distance navigation), task-completion efficiency cannot be overlooked. Some recent methods have considered efficiency in different aspects of TAMP, such as planning task-level optimal behaviors in navigation domains [14], integrating reinforcement learning with symbolic planning in dynamic environments [8], computing safe and efficient plans for urban driving [30], and optimizing robot navigation actions under the uncertainty from motion and sensing [9]. In contrast to those methods that do not have a perception component, GROP visually grounds symbols (about spatial relationships) to probabilistically evaluate action feasibility for task-motion planning.…”

“…On the other hand, this paper is motivated by the latter type of TAMP domains, wherein it is advantageous to incorporate both efficiency and feasibility into the evaluation of plan qualities. Some existing TAMP research incorporates both efficiency and feasibility into task-motion planning [9], [14]. However, those methods evaluate feasibility in a deterministic way, and rely on predefined "state mapping functions" for mapping each symbolic state into feasible poses in continuous space.…”