WorldDreamerV4 targets shared world models for robot swarms and tops RoboCasa, WorldScore
机器之心 · wechat · 2026-07-27
WorldAgents has released WorldDreamerV4, a “shared world-action model” aimed at multi-agent physical AI rather than a single robot.
The paper-style announcement claims a new training recipe built around:
- Multi-agent end-to-end pretraining, where several agents act in the same scene with different views and partial observations.
- Masked Agent Modeling, which hides some agents’ states so the model must infer their current behavior and future actions.
- Shared latent world state, so robots can coordinate through a common evolving world model instead of constant message passing.
The company says the model reached #1 on RoboCasa for robot control and #1 on WorldScore for world generation, and argues that the long-term scaling law for robotics comes from more robots generating more physical-world data.
Its broader thesis is “One World, One Model”: a single world model could connect industrial arms, AGVs, quadrupeds, humanoids, drones, service robots and space robots.
More from Embodied
- Qualcomm robot collapses mid-presentation in an awkward demo fail — iamfakhrealam · 2026-07-27
- Apollo Go starts its first fully driverless trial in Hong Kong — Baidu_Inc · 2026-07-27
- A robotics policy trained on 4 hours of diverse data beats in-domain data — DominiqueCAPaul · 2026-07-27
- World Action Models survey says robotics is moving from reacting to predicting consequences — rohanpaul_ai · 2026-07-27
- AheadForm unveils an ultra-lifelike humanoid robot at WAIC 2026 in Shanghai — Olivier__OG · 2026-07-27
- Stanford AI Lab cites NVIDIA’s Alpamayo platform as a model for open Physical AI — StanfordAILab · 2026-07-27