AI

World Action Modeling with Progressive Visual Planning

Read full article at source ↗

Based on a feed excerpt; full article content has not been acquired. Source evidence retrieved 05/10/2026, 10:18:24 UTC.

Briefings

ProWAM is a progressive world action model that jointly predicts actions and an ordered sequence of sparse visual sub-goals, enabling efficient long-horizon planning by caching sparse features instead of generating full video rollouts. Evaluated on simulation and real-world benchmarks, ProWAM achieved state-of-the-art performance on LIBERO-Plus (85.8%) and randomized RoboTwin (75.7%), with zero-shot real-world success rates reaching 70.0% in novel scenes, significantly outperforming baselines. The model scales by learning sub-goals from action-free videos, reducing computational costs while maintaining robustness against out-of-distribution shifts.

Key points

• ProWAM jointly predicts actions and an ordered sequence of sparse visual sub-goals to enable efficient long-horizon planning without generating full video rollouts.
• On LIBERO-Plus, ProWAM achieved 85.8% success rate, setting a new state-of-the-art benchmark that outperforms the strongest baseline by +35.9%.

Context and limitations

Results are limited to simulation benchmarks and zero-shot real-world experiments; no long-horizon real-world deployment data is provided.

Generated by Bittrees-hosted qwen35-2b sha256:aaf42c8b7c3c on 05/10/2026, 10:19:21 UTC. Based on publisher-provided feed evidence; may not cover the full original article. Check the source for details.

View archived briefing on IPFS
bafybeifxqjamojr2ea6yoxpktrqmdfur4vu7oz63n27unuqmhwnsh3zcsm