SLAP: Shortcut Learning for Abstract Planning

2025-11-02

AI Summary

Traditional Task-and-Motion Planning (TAMP) methods struggle in sparse-reward, continuous state-action domains for long-horizon decision-making, as they rely on hand-crafted abstract actions (e.g., “pick up”, “place”), limiting adaptability and scalability. Method: This paper proposes a shortcut-learning-based automatic option discovery framework that introduces model-free reinforcement learning into abstract planning graphs. It autonomously discovers physically grounded, improvisational options (e.g., “tap”, “shake”) directly in the joint task-motion space—bypassing human priors—and performs hierarchical policy optimization over the learned abstract action graph to co-train high-level planning and low-level motor control. Contribution/Results: Evaluated on four simulated robotic manipulation tasks, our approach reduces average planning path length by over 50% and achieves significantly higher success rates than flat RL, hierarchical RL, and classical TAMP baselines. It improves both generalization across tasks and planning efficiency without domain-specific abstractions.

Technology Category

• Humans and AI: Human-Aware Planning and Behavior Prediction
• Intelligent Robots: Motion and Path Planning
• Planning, Routing, and Scheduling: Learning for Planning and Scheduling

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Responsible Web: Machine-in-the-loop, human agency and autonomy
• Graph Algorithms and Modeling for the Web: Representation, reconstruction, and subgraph or motif discovery in Web-related graphs

Abstract

Long-horizon decision-making with sparse rewards and continuous states and actions remains a fundamental challenge in AI and robotics. Task and motion planning (TAMP) is a model-based framework that addresses this challenge by planning hierarchically with abstract actions (options). These options are manually defined, limiting the agent to behaviors that we as human engineers know how to program (pick, place, move). In this work, we propose Shortcut Learning for Abstract Planning (SLAP), a method that leverages existing TAMP options to automatically discover new ones. Our key idea is to use model-free reinforcement learning (RL) to learn shortcuts in the abstract planning graph induced by the existing options in TAMP. Without any additional assumptions or inputs, shortcut learning leads to shorter solutions than pure planning, and higher task success rates than flat and hierarchical RL. Qualitatively, SLAP discovers dynamic physical improvisations (e.g., slap, wiggle, wipe) that differ significantly from the manually-defined ones. In experiments in four simulated robotic environments, we show that SLAP solves and generalizes to a wide range of tasks, reducing overall plan lengths by over 50% and consistently outperforming planning and RL baselines.

Problem

• Automating discovery of abstract actions in hierarchical planning
• Addressing sparse reward challenges in continuous state-action spaces
• Improving robotic task success through learned shortcut behaviors

Innovation

• Automatically discovers new options using existing TAMP
• Learns shortcuts in abstract graph with model-free RL
• Reduces plan lengths and improves task success rates