Autoregressive Flow Matching for Motion Prediction
Analysis
This paper introduces Autoregressive Flow Matching (ARFM), a novel method for probabilistic modeling of sequential continuous data, specifically targeting motion prediction in human and robot scenarios. It addresses limitations in existing approaches by drawing inspiration from video generation techniques and demonstrating improved performance on downstream tasks. The development of new benchmarks for evaluation is also a key contribution.
Key Takeaways
- •Proposes Autoregressive Flow Matching (ARFM) for probabilistic modeling of sequential continuous data.
- •Applies ARFM to motion prediction in human and robot scenarios.
- •Demonstrates improved performance on downstream tasks by conditioning on predicted future tracks.
- •Develops new benchmarks for evaluating motion prediction models.
- •Inspired by scaling of video generation techniques.
Reference
“ARFM is able to predict complex motions, and we demonstrate that conditioning robot action prediction and human motion prediction on predicted future tracks can significantly improve downstream task performance.”