OSWAM: An Effective World Action Model with One-Step Action Generation

Researchers introduced OSWAM, a World Action Model (WAM) that generates actions in a single step rather than the tens of denoising steps required by existing direct-policy WAMs, according to the paper's description. The work targets efficient inference for autonomous driving and embodied control, where prior WAMs decoupled action generation from explicit future visual prediction but still needed many denoising steps to reach competitive action quality.

Direct-policy World Action Models WAM have demonstrated the benefits of decoupling action generation from explicit future visual prediction for efficient inference in autonomous driving and embodied control. However, they still require tens of denoising steps to achieve competitive action quality,