Challenges the necessity of discrete action tokenizers in robotics by using a continuous, single-stage flow matching policy.
March 31, 2026
Original Paper
HiFlow: Tokenization-Free Scale-Wise Autoregressive Policy Learning via Flow Matching
arXiv · 2603.27281
The Takeaway
It eliminates the quantization error and multi-stage training pipelines inherent in transformer-based robot policies that treat low-dimensional actions like high-dimensional pixels, outperforming existing diffusion and tokenization methods.
From the abstract
Coarse-to-fine autoregressive modeling has recently shown strong promise for visuomotor policy learning, combining the inference efficiency of autoregressive methods with the global trajectory coherence of diffusion-based policies. However, existing approaches rely on discrete action tokenizers that map continuous action sequences to codebook indices, a design inherited from image generation where learned compression is necessary for high-dimensional pixel data. We observe that robot actions are