WAIC panel says world models still lack consensus on routes, metrics, and deployment
量子位 · wechat · 2026-07-21
WAIC forum surfaces the biggest splits in world-model research
This long article summarizes a WAIC roundtable with six companies working on world models and embodied AI, including speakers from Daxia Robot, Ant Lingbo, Jijia Vision, Wujie Power, Agibot, and Subitron.
The major disagreements
- Technical route: pixel generation, latent-space JEPA, or hybrid approaches.
- What to optimize: strict physical accuracy versus only physical plausibility.
- Evaluation: long-horizon consistency versus fast inference that fits real-time decision windows.
What participants think will converge first
- A unified representation for vision, language, and action.
- Fusion of video generation, latent reasoning, and explicit 3D memory.
- Touch as a required modality for robots.
Deployment is the real test
The discussion stresses that demos are easy, but deployment is hard:
- real-world error rates become expensive at scale,
- robots must handle random disruptions and human interference,
- and edge/on-device inference matters because cloud latency can break control loops.
The broader takeaway
The host concludes that disagreement is still the raw material for papers, while overlap is what eventually becomes industry consensus.
More from Embodied
- FLASH: GPU-native deformable sim learns cloth folding in minutes on one RTX 5090 — ChongZzZhang · 2026-10-26
- Luke Wroblewski spots it: iPhone Duo's corner radius mirrors the original iPhone — LukeW · 2026-10-26
- Humanoids shattered forecasts: half-marathon record in 2026, 100m faster than Bolt — ben_j_todd · 2026-10-26
- Humanoid walking is continuous correction — watch recovery, not locomotion, in demos — Olivier__OG · 2026-10-26
- CircuitHub unveils The Grid, a factory-scale robot making PCBs in the US twice as fast as Shenzhen — garrytan · 2026-10-26
- Shanghai AI Lab-led team releases 100-hour egocentric wearable dataset for humanoid training — ChuckDBrooks · 2026-10-26