Efficient Diffusion Policy Fine Tuning with Latent Noise Representation Bridging
Organizations: The Australian National University. · Maincode.
Abstract
Pre-trained diffusion and flow policies have emerged as powerful backbones for robotics, yet adapting them to novel data via reinforcement learning (RL) remains computationally expensive, sample inefficient and lacks theoretical guarantees. Prior approaches that fine-tune the initial noise bypass costly backpropagation through time but struggle with the "noise aliasing" problem: there does not exista natural link between the noise space value and the action space value. This paper resolves these challenges by introducing the Latent Noise Linear Markov Decision Process (LNL-MDP). We derive the Bridge Equation for LNL-MDPs which eliminates the noise aliasing problem by aligning values with a shared representation between the noise and the action space. Furthermore, our LNL-MDP fine tuning framework falls under the FineTuneRL theory with near-optimal guarantee. Leveraging these insights, we propose Latent Representation Bridging for Diffusion (LARBRID), an algorithm that efficiently learns the representation to optimize the initial distribution that steers the pre-trained model to act optimally in the environment. It yields up to 2-3 performance increases on challenging control problems across a diverse suite of robotic locomotion and manipulation tasks. Code can be found at: github.com/Maincode-Research/LARBRID.git.
Figures & tables
| Method | ant-large | ant-giant | hum-med | hum-large | soccer | cube-s | cube-d | scene | pz-3x3 | pz-4x4 | visual |
|---|---|---|---|---|---|---|---|---|---|---|---|
| IQL | |||||||||||
| REBRAC | |||||||||||
| IDQL | |||||||||||
| SRPO | |||||||||||
| CAC | |||||||||||
| FAWAC |
| Method | cube-double | scene | puzzle-4x4 | hum-med | ant soccer |
|---|---|---|---|---|---|
| IQL | |||||
| REBRAC | |||||
| Cal-QL | |||||
| RLPD | |||||
| IFQL | |||||
| FQL |
Appendix figures & tables7 assets
Supplementary material from the paper’s appendix.
Appendix
| Task | IQL | ReBRAC | FBRAC | IFQL | FQL | LARBRID-I |
|---|---|---|---|---|---|---|
| visual-cube-single-play-singletask-task1-v0 | ||||||
| visual-cube-double-play-singletask-task1-v0 | ||||||
| visual-scene-play-singletask-task1-v0 | ||||||
| visual-puzzle-3x3-play-singletask-task1-v0 | ||||||
| visual-puzzle-4x4-play-singletask-task1-v0 |
| Hyperparameters | Value |
| representation learning rates | 1e-4 |
| other learning rates | 3e-4 |
| Target update rate ( ) | 5e-3 |
| Target entropy (online) | 0 |
| Target entropy ( ) | |
| Critic and actor #. of hidden layers | 2 |
Robot Policy Learning
Robot policy learning develops controllers that map sensory observations, task goals, and interaction history to actions for embodied robots.
See all 405 papers