Abstract

Imitation learning offers a promising approach in robotics by enabling systems to learn directly from data without requiring explicit models, simulations, or detailed task definitions. During inference, actions are sampled from the learned distribution and executed on the robot. However, sampled actions may fail for various reasons, and simply repeating the sampling step until a successful action is obtained can be inefficient. In this work, we propose an enhanced sampling strategy that refines the sampling distribution to avoid previously unsuccessful actions. We demonstrate that by solely utilizing data from successful executions, our method can infer recovery actions without the need for additional simulation or exploratory behavior. Furthermore, we leverage the concept of diffusion model decomposition to break down the primary problem—which may require long-horizon history to manage failures—into multiple smaller, more manageable sub-problems in learning, data collection, and inference, thereby enabling the system to adapt to variable failure counts. Our approach yields a low-level controller that dynamically adjusts its sampling space to improve efficiency when prior samples fall short. We validate our method across several tasks, including door opening with unknown directions, object manipulation, and button-searching scenarios, demonstrating that our approach outperforms traditional baselines.

Bibtex reference

@inproceedings{Razmjoo25IROS,
	author={Razmjoo, A. and Calinon, S. and Gienger, M. and Zhang, F.},
	title={{CCDP}: Composition of Conditional Diffusion Policies with Guided Sampling},
	booktitle={Proc.\ {IEEE/RSJ} Intl Conf.\ on Intelligent Robots and Systems ({IROS})},
	year={2025}
}

Go back to the list of publications