Efficient Diffusion Policy Fine Tuning with Latent Noise Representation Bridging
Efficient Diffusion Policy Fine Tuning with Latent Noise Representation Bridging
This research addresses the computational expense and sample inefficiency of adapting pre-trained diffusion policies to novel data via reinforcement learning. The authors introduce the Latent Noise Linear Markov Decision Process (LNL-MDP) and derive a Bridge Equation to eliminate noise aliasing by aligning noise space values with action space values. They propose LARBRID, an algorithm that efficiently learns representations to optimize initial distributions for pre-trained models. Experiments show up to 2-3x performance increases on challenging control problems across diverse robotic locomotion and manipulation tasks. The code is available at github.com/Maincode-Research/LARBRID.git.
The paper introduces LARBRID, an algorithm that uses Latent Noise Linear Markov Decision Process (LNL-MDP) to resolve noise aliasing in diffusion policy fine-tuning, achieving up to 2-3x performance gains on robotic locomotion and manipulation tasks with near-optimal theoretical guarantees under FineTuneRL.
Source: arXiv Robotics — research abstracts · Read original article ↗
Loading article text…
What the source reports
Publisher-reported claims, with original evidence. These results have not been independently verified by RoboSignal.
- code: Reported pending Open source S5
- paper: not_reported
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
cross a diverse suite of robotic locomotion and manipulation tasks. Code can be found at: github.com/Maincode-Research/LARBRID.git.
Open source S5
Source：arXiv Robotics — research abstracts · arxiv.org