The authors cast open-world embodied navigation-planning as a testbed for EWAs, using camera translation and rotation as universal actions.
Large-scale dataset and world-model RL improve embodied navigation in open worlds
The authors present EmbodiedWorld-200K, over 200K annotated navigation-planning samples, and show that interactive reinforcement learning inside a video world model boosts generalization to unseen scenes.
Chinese Tech
X. Feng · J. Tang · Z. Lin · L. Chen · S. Shao · Q. Lu · +1 more
Tencent Hunyuan
Research Digest··2 min read
The authors propose Embodied World Agents (EWAs), agents that perceive and act across diverse open-world 3D environments, and release EmbodiedWorld-200K, a dataset of over 200K video samples with multi-level annotations.
§