Iterative Compositional Data Generation for Robot Control

2025-12-11

AI Summary

Data collection for robotic manipulation tasks involving combinations of multiple objects, robots, and environments is costly and yields poor generalization. Method: We propose a semantic-composable Diffusion Transformer that decouples state transitions into four semantic components—robot, object, obstacle, and goal—and models their interactions via attention mechanisms. We further introduce an iterative self-improving synthetic data paradigm: zero-shot generation → offline RL validation (using CQL/BCQ) → feedback-augmented knowledge distillation. Contribution/Results: This work establishes the first interpretable and generalizable semantic compositional representation for robot control. It achieves near 100% zero-shot policy success on unseen task compositions, significantly outperforming monolithic models and hand-coded compositional baselines. The synthesized data directly enables end-to-end policy learning without real-world fine-tuning.

Technology Category

• Intelligent Robots: Manipulation
• Computer Vision: Diffusion Models for Vision
• Natural Language Processing: Code Generation / Program Synthesis from Natural Language

Application Category

• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Economics, Online Markets and Human Computation: Humans versus LLMs for data annotation and labeling
• Responsible Web: Machine-in-the-loop, human agency and autonomy

Abstract

Collecting robotic manipulation data is expensive, making it impractical to acquire demonstrations for the combinatorially large space of tasks that arise in multi-object, multi-robot, and multi-environment settings. While recent generative models can synthesize useful data for individual tasks, they do not exploit the compositional structure of robotic domains and struggle to generalize to unseen task combinations. We propose a semantic compositional diffusion transformer that factorizes transitions into robot-, object-, obstacle-, and objective-specific components and learns their interactions through attention. Once trained on a limited subset of tasks, we show that our model can zero-shot generate high-quality transitions from which we can learn control policies for unseen task combinations. Then, we introduce an iterative self-improvement procedure in which synthetic data is validated via offline reinforcement learning and incorporated into subsequent training rounds. Our approach substantially improves zero-shot performance over monolithic and hard-coded compositional baselines, ultimately solving nearly all held-out tasks and demonstrating the emergence of meaningful compositional structure in the learned representations.

Problem

• Generates robotic transitions for unseen task combinations
• Overcomes data scarcity in multi-object multi-robot settings
• Enables zero-shot policy learning via compositional synthesis

Innovation

• Semantic compositional diffusion transformer factorizes transitions
• Zero-shot generation of high-quality transitions for unseen tasks
• Iterative self-improvement with offline reinforcement learning validation