BAAI's Orca Explores Next-State Prediction
新智元 · wechat · 2026-07-08
The BAAI Wujie RoboBrainOrca team released a technical report, "Orca: The World is in Your Mind," exploring a shift from predicting the "next token/frame/action" to predicting the "next state" to learn more unified world representations. The article emphasizes that it is not merely a chatbot, video generator, or robotic policy model; instead, it first learns how world states change and then decodes this representation into capabilities for understanding, prediction, and action.
The post also notes that Orca has sparked discussions within the research community around world models, multimodal representations, and state modeling, landing on the HuggingFaceDailyPapers monthly chart. The core idea is that text, images, video, and actions can all be viewed as different projections of the same real world. If a model can first learn transferable world representations and then integrate minimal action data, it could potentially reduce robot training costs and enhance generalization.
Related event: BAAI Releases Orca World Model Technical Report(2 posts)→
More from Embodied
- DONIC hits 86.79% on bimanual dexterous tracking, crushing prior SOTA's 9.43% — liuziwei7 · 2026-10-27
- Bedrock Robotics raises $600M as its fully autonomous excavators do paid work on live job sites — Sethwinterroth · 2026-10-27
- AW-0 hits 80.8% on RoboTwin tasks, beating strongest baseline by 16 points — trevordarrell · 2026-10-27
- AW-0's code stream fixes world model prediction: picks mango where baseline grabs apple — trevordarrell · 2026-10-27
- AW-0 runs at 230ms on a single RTX 5090, 2.4x faster than autoregressive code planning — trevordarrell · 2026-10-27
- AW-0 debuts: one generative world action model plans in code, predicts, and controls robots — trevordarrell · 2026-10-27