Generative Predictive Control: Flow Matching Policies for Dynamic and Difficult-to-Demonstrate Tasks

TL;DR

This paper introduces generative predictive control, a new framework that leverages flow matching policies for fast, dynamic tasks, overcoming the limitations of requiring demonstrations and slow task execution.

Contribution

It proposes a supervised learning approach for fast, dynamic tasks using flow matching policies, enabling high-frequency feedback without expert demonstrations.

Findings

Effective in tasks with fast dynamics and complex behaviors

Maintains temporal consistency during inference

Enables high-frequency feedback for dynamic control

Abstract

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.