Event
Proposing the WorldScape Policy 2.0 world action model
1 reports · 1 sources
Haisheng Su et al. proposed WorldScape Policy 2.0 on arXiv, a controllable world action model with reasoning-enhanced long short-term memory, and constructed an event-level embodied pretraining dataset, ManipEvent-5M, containing nearly 5 million event segments with aligned action trajectories. The model's causal short-term visual memory uses recent observations as DiT prefill to preserve local interaction dynamics, while the long-term event memory organizes historical VLM outputs into global history, locally active, and event-boundary representations for progress-aware retrieval and generation of implicit subgoal conditions. The paper states that in experiments on simulation and real platforms it demonstrates long-horizon autonomous planning, fine-grained instruction following, and contextual adaptation capabilities.
Report timeline
- WorldScape Policy 2.0: Steerable World Action Modeling via Reasoning-Augmented Memory and In-Context Learning
The paper proposes WorldScape Policy 2.0, a controllable World Action Model (WAM) with reasoning-augmented long short-term memory. Its causal short-term visual memory supplies recent observations as DiT prefill to preserve local interaction dynamics, while its long-term event memory organizes historical VLM outputs into global-history, local-active, and event-boundary representations for progress-aware retrieval and to generate implicit subgoal conditions.
arXiv cs.RO ·