Learning Real-World Action-Video Dynamics with Heterogeneous Masked Autoregression
This work addresses the problem of generating high-quality video data and evaluation for scaling robot learning, representing an incremental improvement in domain-specific robotic video generation.
The paper tackles the challenge of modeling action-video dynamics for robotics by proposing Heterogeneous Masked Autoregression (HMA), which achieves better visual fidelity and controllability than previous models with 15 times faster speed in real-world applications.
We propose Heterogeneous Masked Autoregression (HMA) for modeling action-video dynamics to generate high-quality data and evaluation in scaling robot learning. Building interactive video world models and policies for robotics is difficult due to the challenge of handling diverse settings while maintaining computational efficiency to run in real time. HMA uses heterogeneous pre-training from observations and action sequences across different robotic embodiments, domains, and tasks. HMA uses masked autoregression to generate quantized or soft tokens for video predictions. \ourshort achieves better visual fidelity and controllability than the previous robotic video generation models with 15 times faster speed in the real world. After post-training, this model can be used as a video simulator from low-level action inputs for evaluating policies and generating synthetic data. See this link https://liruiw.github.io/hma for more information.