Q-learning method speeds up fine-tuning of complex flow policies
Q-Learning with Scalar Adjoint Matching
Machine LearningArtificial IntelligenceRobotics
Summary
Fine-tuning AI systems that generate actions through multiple steps is difficult and slow because the usual methods require expensive calculations at every step. The authors found a way to simplify this by using a math trick that treats the problem as one scalar value instead of many complex vectors. This lets them update these AI systems much more efficiently, making it easier to improve performance on challenging tasks. They tested their method on both standard benchmarks and a robot system, showing it works better than previous approaches.
What this means in practice
- •For robotics engineers: Fine-tune pretrained robot policies more efficiently for complex manipulation tasks using scalar adjoint Q-learning.
- •For reinforcement learning developers: Improve off-policy training of multi-step flow policies in challenging simulated environments with a faster scalar adjoint method.
Authors
Yonghoon Dong, Minsung Yoon, Jaehyuk Kim, Jungwoo Park, Changyeon Kim, Jinwoo Shin
Abstract
Flow policies capture rich and diverse action distributions, and fine-tuning them with off-policy RL to improve beyond the demonstrations has drawn growing interest. However, fine-tuning a flow policy against a learned value function is not trivial, because the policy generates its action over many flow steps. Adjoint matching offers a principled way to update the flow model itself by propagating value information from the final action back to each flow step, but it requires a vector--Jacobian product through the policy at every step, a cost that grows with the number of flow steps and the policy size. We observe that the batch-averaged velocity Jacobian of pretrained flow policies concentrates on its diagonal. Motivated by this finding, we derive a closed-form scalar adjoint that scales the value gradient at the final action by the flow time, eliminating the per-step vector--Jacobian products. We further find that controlling the critic's value at policy-generated actions is particularly important under the scalar adjoint. Based on these findings, we propose Q-learning with Scalar Adjoint Matching (SQAM), which combines the scalar adjoint with a value penalty at those actions. SQAM's gains concentrate on the four hardest OGBench domains, where its success rate exceeds that of the strongest baseline in each domain by 18 to 35 percentage points. To test whether SQAM extends to large pretrained policies, we also fine-tune a vision-language-action policy on a real bimanual robot. SQAM improves over supervised fine-tuning on all three tasks.