[Paper Review] Task Planning in Robotics: an Empirical Comparison of PDDL-based and ASP-based Systems

Yuqian Jiang, Shiqi Zhang+2 more|arXiv (Cornell University)|Apr 23, 2018

Logic, Reasoning, and Knowledge30 references4 citations

TL;DR

This paper empirically compares PDDL-based and ASP-based task planners in robotics, using identical domain encodings across three benchmarks. It finds that PDDL-based planners outperform on long-plan problems, while ASP-based planners excel in large-object or complex-reasoning domains, offering practical guidance for planner selection in robotic task planning.

ABSTRACT

Robots need task planning algorithms to sequence actions toward accomplishing goals that are impossible through individual actions. Off-the-shelf task planners can be used by intelligent robotics practitioners to solve a variety of planning problems. However, many different planners exist, each with different strengths and weaknesses, and there are no general rules for which planner would be best to apply to a given problem. In this article, we empirically compare the performance of state-of-the-art planners that use either the Planning Domain Description Language (PDDL), or Answer Set Programming (ASP) as the underlying action language. PDDL is designed for task planning, and PDDL-based planners are widely used for a variety of planning problems. ASP is designed for knowledge-intensive reasoning, but can also be used for solving task planning problems. Given domain encodings that are as similar as possible, we find that PDDL-based planners perform better on problems with longer solutions, and ASP-based planners are better on tasks with a large number of objects or in which complex reasoning is required to reason about action preconditions and effects. The resulting analysis can inform selection among general purpose planning systems for particular robot task planning domains.

Motivation & Objective

• To guide roboticists in selecting between PDDL-based and ASP-based task planners based on domain characteristics.
• To evaluate the performance of state-of-the-art PDDL and ASP planners under comparable domain encodings.
• To identify which planner type performs better under varying domain properties such as plan length, number of objects, and reasoning complexity.
• To provide actionable insights for practitioners choosing general-purpose planning systems in robotics.

Proposed method

• The study uses three benchmark domains: Blocks World, Hiking, and a variant of Robot Navigation from the IPC and prior robotics work.
• Domains are encoded in both PDDL and ASP using as similar representations as possible to ensure fair comparison.
• The FastDownward planner (FDSS-1 and LAMA-2011) is used for PDDL-based planning, and Clingo for ASP-based planning.
• Experiments vary key domain properties: plan length, number of objects, and complexity of action preconditions and effects.
• Planning time is measured across multiple runs on a heterogeneous HTC cluster to assess consistency across hardware.
• Statistical analysis is performed to confirm trends, with low standard deviations indicating robustness across different computing platforms.

Experimental results

Research questions

• RQ1 Does the PDDL-based planner outperform the ASP-based planner on problems requiring long sequences of actions?
• RQ2 Does the ASP-based planner perform better on domains with a large number of objects, despite higher grounding costs?
• RQ3 Does the ASP-based planner show superior performance on tasks requiring complex reasoning about action preconditions and effects?
• RQ4 How do variations in domain size and complexity affect the relative performance of PDDL and ASP planners?

Key findings

• PDDL-based planners significantly outperform ASP-based planners on problems with long solutions, particularly in the Hiking and Blocks World domains.
• ASP-based planners demonstrate superior scalability with increasing numbers of domain objects, especially when the number of grounded actions does not explode.
• In the Robot Navigation domain, which requires complex reasoning about action preconditions and effects, ASP-based planners are considerably faster than PDDL-based planners for shorter plans.
• The performance gap between the two paradigms is consistent across diverse hardware configurations, indicating robust trends independent of machine-specific factors.
• Clingo’s advantage in large domains is partially offset by higher grounding time, suggesting a trade-off between grounding and solving efficiency.
• The results support the hypothesis that PDDL is better suited for long-plan problems, while ASP is better for complex or large-scale reasoning tasks.