W2A: Distilling World Priors into Action Models for Real-Time Robotic Control
Overview
This research introduces W2A, a two-stage distillation framework designed to enable real-time robotic manipulation by transferring intermediate denoising representations from a video diffusion model into an action model during training. The approach discards the video model at test time to reduce computational overhead and latency. Experiments in both simulation and real-world settings show that W2A achieves a 95.9% average success rate on the LIBERO benchmark and approximately 21x faster inference compared to the representative World Action Model (WAM) baseline.
Generated from attributed reports · Updated 1 hours ago
Event evidence and corrections
0 attributed source owners. Ownership does not establish independent confirmation. Quantities are reported separately and are never added together.
Reported quantity · inference speed improvement: 21 other · Basis not reported
“fective supervision for action generation. This makes it possible to transfer such information effectively into an action model. Extensive experiments in both simulation and real-world settings demonstrate competitive success rates while significantly improving inference efficiency. On LIBERO, W2A achieves a 95.9% average success rate. More importantly, W2A achieves about 21x faster inference than the representative”
Exact source · revision 1Source owner not reported
Report timeline
Follow attributed reports and material updates.
- arXiv Robotics — research abstractsSignalW2A: Distilling World Priors into Action Models for Real-Time Robotic Control
This research introduces W2A, a two-stage distillation framework designed to enable real-time robotic manipulation by transferring intermediate denoising representations from a video diffusion model into an action model during training. The approach discards the video model at test time to reduce computational overhead and latency. Experiments in both simulation and real-world settings show that W2A achieves a 95.9% average success rate on the LIBERO benchmark and approximately 21x faster inference compared to the representative World Action Model (WAM) baseline.
Event coverage history
There is not enough continuous observation data to show a trend.
Timezone · UTC
Article dates follow your selected timezone. Briefing editions use Hong Kong time (UTC+8).