PRAG: Procedural Action Generator

2025-07-12

AI Summary

Existing approaches lack scalable and verifiable methods for programmatically generating multi-step, high-contact robotic manipulation tasks. Method: This paper proposes a symbolic-physical co-verification framework that enables users to define atomic actions, objects, and spatial predicates; it enforces three complementary constraints—logical consistency checking, object-predicate compatibility analysis, and simulation-based solvability verification—to ensure semantic and physical feasibility of generated tasks. Contribution/Results: The framework generates structured task sequences up to 15 steps long, producing dense semantic task sets annotated with subgoal rewards and state labels—directly usable for reinforcement learning training or high-quality benchmark construction. Experiments yield over one million unique, verifiably solvable tasks, supporting semantic similarity computation and efficient downstream policy learning. The approach significantly improves controllability, formal verifiability, and practical utility in robotic task generation.

Technology Category

• Intelligent Robots: Manipulation
• Humans and AI: Human-Aware Planning and Behavior Prediction
• Constraint Satisfaction and Optimization: Satisfiability Modulo Theories

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Economics, Online Markets and Human Computation: LLM based quality controls for crowd work

Abstract

We present a novel approach for the procedural construction of multi-step contact-rich manipulation tasks in robotics. Our generator takes as input user-defined sets of atomic actions, objects, and spatial predicates and outputs solvable tasks of a given length for the selected robotic environment. The generator produces solvable tasks by constraining all possible (nonsolvable) combinations by symbolic and physical validation. The symbolic validation checks each generated sequence for logical and operational consistency, and also the suitability of object-predicate relations. Physical validation checks whether tasks can be solved in the selected robotic environment. Only the tasks that passed both validators are retained. The output from the generator can be directly interfaced with any existing framework for training robotic manipulation tasks, or it can be stored as a dataset of curated robotic tasks with detailed information about each task. This is beneficial for RL training as there are dense reward functions and initial and goal states paired with each subgoal. It allows the user to measure the semantic similarity of all generated tasks. We tested our generator on sequences of up to 15 actions resulting in millions of unique solvable multi-step tasks.

Problem

• Generates solvable multi-step robotic manipulation tasks
• Validates tasks via symbolic and physical constraints
• Outputs tasks compatible with RL training frameworks

Innovation

• Procedural generation of solvable robotic tasks
• Symbolic and physical validation for task feasibility
• Direct interface with RL training frameworks