PAC-DP: PAC-Bayesian Diffusion Policy Learning
2026-07-27 • Robotics
Robotics
AI summaryⓘ
The authors studied Diffusion Policies (DPs), which help robots perform complex tasks but struggle to generalize well when trained with limited data. They introduced PAC-DP, a new method that treats DPs like Bayesian neural networks and adds a special mathematical tool (a PAC-Bayes bound) to better guide learning. This approach adds a regularization term that improves learning without much extra training time. Experiments showed their method helps robots perform better, especially when data is scarce or tasks are hard. Overall, the authors provide a theoretically sound way to improve robot learning with fewer data.
Diffusion PoliciesRobotic ManipulationBayesian Neural NetworksPAC-Bayes BoundGeneralizationDenoising ObjectiveKullback-Leibler DivergenceRegularizationVariational Negative Log-LikelihoodLow-Data Regimes
Authors
Mohammad Hasan Yeganegi, Dian Yu, Andrea Del Prete, Majid Khadiv, Matteo Saveriano
Abstract
Diffusion Policies (DPs) are able to perform complex manipulation tasks. However, DPs are typically trained by minimizing a denoising objective, which provides limited control over generalization in the finite-data regimes common in robotics. In this letter, we propose PAC-DP, an approach that increases the performance of DPs in robotic manipulation tasks. By modeling the DP as a Bayesian neural network, and defining a PAC-Bayes generalization bound, we derive a novel training objective that augments the standard denoising loss with a Kullback-Leibler divergence regularizer between the posterior and prior parameter distributions. From the theoretical perspective, our approach provides a principled approach to regularize the training of DPs without significantly increasing the training time. From the practical point of view, experimental results demonstrate improved denoising performance, lower variational negative log-likelihood, and higher success rates across multiple robotic manipulation benchmarks. Crucially, the largest improvements are observed in low-data training regimes and complex tasks, establishing PAC-DP as a theoretically grounded framework for robot policy learning.