Generative Predictive Control: Flow Matching Policies for Dynamic and Difficult-to-Demonstrate Tasks
This addresses the problem of enabling robots to perform dynamic tasks without expert demonstrations, offering a complementary approach to behavior cloning for robotics researchers, though it appears incremental as it builds on existing generative methods.
The paper tackles the limitations of generative control policies that require expert demonstrations and are limited to slow tasks by introducing generative predictive control, a supervised learning framework for fast-dynamic tasks that are easy to simulate but hard to demonstrate, achieving high-frequency feedback through warm-started flow-matching policies.
Generative control policies have recently unlocked major progress in robotics. These methods produce action sequences via diffusion or flow matching, with training data provided by demonstrations. But existing methods come with two key limitations: they require expert demonstrations, which can be difficult to obtain, and they are limited to relatively slow, quasi-static tasks. In this paper, we leverage a tight connection between sampling-based predictive control and generative modeling to address each of these issues. In particular, we introduce generative predictive control, a supervised learning framework for tasks with fast dynamics that are easy to simulate but difficult to demonstrate. We then show how trained flow-matching policies can be warm-started at inference time, maintaining temporal consistency and enabling high-frequency feedback. We believe that generative predictive control offers a complementary approach to existing behavior cloning methods, and hope that it paves the way toward generalist policies that extend beyond quasi-static demonstration-oriented tasks.