Improving Generative Behavior Cloning via Self-Guidance and Adaptive Chunking

2025-10-14

AI Summary

Generative Behavior Cloning (GBC) enables efficient multi-task robotic learning but suffers from two key limitations: stochastic diffusion policies often yield erroneous action samples, and open-loop control incurs latency, hindering adaptation to dynamic environments. To address these, we propose a self-guided and adaptive chunking framework: (1) a self-guidance mechanism that fuses historical observations to enhance future state anticipation; and (2) an adaptive chunking strategy that dynamically adjusts the temporal granularity of action sequence updates, optimizing the trade-off between reactivity and temporal consistency. Our approach builds upon diffusion models without requiring additional supervision or online replanning. Evaluated on both simulation and real-robot multi-task manipulation benchmarks, it achieves significant improvements in task success rate (+12.3%) and cross-task generalization. The code is publicly available.

Technology Category

• Intelligent Robots: Manipulation
• Humans and AI: Human-Aware Planning and Behavior Prediction
• Machine Learning: Adversarial Learning & Robustness

Application Category

• Economics, Online Markets and Human Computation: Economic ramifications for generative AI infrastructure and applications
• Responsible Web: Machine-in-the-loop, human agency and autonomy
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Generative Behavior Cloning (GBC) is a simple yet effective framework for robot learning, particularly in multi-task settings. Recent GBC methods often employ diffusion policies with open-loop (OL) control, where actions are generated via a diffusion process and executed in multi-step chunks without replanning. While this approach has demonstrated strong success rates and generalization, its inherent stochasticity can result in erroneous action sampling, occasionally leading to unexpected task failures. Moreover, OL control suffers from delayed responses, which can degrade performance in noisy or dynamic environments. To address these limitations, we propose two novel techniques to enhance the consistency and reactivity of diffusion policies: (1) self-guidance, which improves action fidelity by leveraging past observations and implicitly promoting future-aware behavior; and (2) adaptive chunking, which selectively updates action sequences when the benefits of reactivity outweigh the need for temporal consistency. Extensive experiments show that our approach substantially improves GBC performance across a wide range of simulated and real-world robotic manipulation tasks. Our code is available at https://github.com/junhyukso/SGAC

Problem

• Addresses erroneous action sampling in diffusion policies
• Mitigates delayed responses in open-loop robot control
• Enhances consistency and reactivity of generative behavior cloning

Innovation

• Self-guidance improves action fidelity using past observations
• Adaptive chunking selectively updates action sequences dynamically
• Enhanced diffusion policies increase consistency and reactivity