Efficient Symbolic Planning with Views

TL;DR

The paper addresses the challenge of generalizing robotic planning across devices and environments by replacing coarse, compound attributes with an atomic, elementary attribute space. It introduces view-based planning (VBP), which solves a planning problem $T=(\mathcal{A}, s, g)$ in a sequence of views $V=(\mathcal{P}^1, \ldots, \mathcal{P}^N)$ with increasing predicate sets and $\mathcal{P}^N=\mathcal{P}$, exploiting independence between attribute groups to reduce search complexity. Per view, FILTER reduces the problem to the current predicates, SOLVE finds a plan, and MODIFY grounds and preserves actions for subsequent views, ensuring the final view yields a solution to the original problem. Empirical results on an example domain show that VBP can produce low-cost plans quickly and detect uns solvable goals early, outperforming standalone satisficing or optimal planners in interactive scenarios, while still allowing more optimal subproblem solutions when desired. Limitations include the need for appropriate predicate-group independence and the lack of backtracking across views; future work includes automated view construction and human-guided guidance to favor common solutions.

Abstract

Robotic planning systems model spatial relations in detail as these are needed for manipulation tasks. In contrast to this, other physical attributes of objects and the effect of devices are usually oversimplified and expressed by abstract compound attributes. This limits the ability of planners to find alternative solutions. We propose to break these compound attributes down into a shared set of elementary attributes. This strongly facilitates generalization between different tasks and environments and thus helps to find innovative solutions. On the down-side, this generalization comes with an increased complexity of the solution space. Therefore, as the main contribution of the paper, we propose a method that splits the planning problem into a sequence of views, where in each view only an increasing subset of attributes is considered. We show that this view-based strategy offers a good compromise between planning speed and quality of the found plan, and discuss its general applicability and limitations.

Figures (2)

• Figure 1: Snapshot of the simulation environment. For clarity, only some symbolically modelled objects and devices are omitted. The robot can move in one dimension along the table and manipulate objects with its two arms.
• Figure 2: The view-based plan for goal 3 (mint iced tea). In each view the used predicate groups are shown on top, and the found plan as list of actions below. Each action is given as name -- list of grounded parameters (objects) and darker shaded actions are introduced in the corresponding view.