We describe anchored angelic A* (AAA*), an algorithm for efficient and asymptotically near-optimal planning for systems governed by piecewise-analytic differential constraints. This class of systems includes many task and motion planning domains, such as quasi-static manipulation planning. AAA* uses an abstract representation of the problem domain to extract upper and lower bounds on the cost o...

The use of a policy and heuristic function for guiding search can be quite effective in adversarial problems, as demonstrated by AlphaGo its successors, which are based on the PUCT algorithm. While also used to solve single-agent deterministic it lacks guarantees effort computationally inefficient practice. Combining A* algorithm with learned tends work better these domains, but variants do not...

When given several problems to solve in some domain, a standard reinforcement learner learns an optimal policy from scratch for each problem. If the domain has particular characteristics that are goal and problem independent, the learner might be able to take advantage of previously solved problems. Unfortunately, it is generally infeasible to directly apply a learned policy to new problems. Th...

Heuristic search is a leading approach to domain-independent planning. For cost-optimal planning, however, existing admissible heuristics are generally too weak to effectively guide the search. Pattern database heuristics (PDBs), which are based on abstractions of the search space, are currently one of the most promising approaches to developing better admissible heuristics. The informedness of...

We present a nondeterministic model of computation based on reversing edge directions in weighted directed graphs with minimum in-flow constraints on vertices. Deciding whether this simple graph model can be manipulated in order to reverse the direction of a particular edge is shown to be PSPACEcomplete by a reduction from Quantified Boolean Formulas. We prove this result in a variety of specia...

We present a nondeterministic model of computation based on reversing edge directions in weighted directed graphs with minimum in-flow constraints on vertices. Deciding whether this simple graph model can be manipulated in order to reverse the direction of a particular edge is shown to be PSPACEcomplete by a reduction from Quantified Boolean Formulas. We prove this result in a variety of specia...

The planner GRAPHPLAN is based on an efficient propagation of reachability information which then effectively guides a search for a valid plan. We propose a framework in which a broader class of information, including the original reachability information, can be propagated in the plan graph in polynomial time. As an example, we exhibit an algorithm for propagating “landmark” information, where...

Planning problems are among the most important and well-studied problems in artificial intelligence. They are most typically solved by tree search algorithms that simulate ahead into the future, evaluate future states, and back-up those evaluations to the root of a search tree. Among these algorithms, Monte-Carlo tree search (MCTS) is one of the most general, powerful and widely used. A typical...

This paper proves that push-pull block puzzles in 3D are PSPACE-complete to solve, and push-pull block puzzles in 2D with thin walls are NP-hard to solve, settling an open question [ZR11]. Pushpull block puzzles are a type of recreational motion planning problem, similar to Sokoban, that involve moving a ‘robot’ on a square grid with 1 × 1 obstacles. The obstacles cannot be traversed by the rob...