NVIDIA: Embodied AI Also Benefits from Context Scaling
dair_ai · x · 2026-07-17
A new NVIDIA paper brings context scaling to embodied AI: foundational robot models shouldn't only look at single steps or short histories, as real-world assembly tasks often last minutes and involve multi-stage decision-making.
The paper introduces RoboTTT, which expands context length by three orders of magnitude over current policies without increasing inference latency. The authors report:
- An overall performance improvement of 87% over single-step baselines on real robot manipulation
- The ability to complete a 5-minute, 10-step assembly task, which baselines failed to finish
- The 8K timestep model outperforming the 1K pre-trained version of the same model
- The first observation that longer pre-training context consistently improves closed-loop performance
The authors emphasize that "context length" itself could be a critical scaling axis in embodied intelligence.
Related event: RoboTTT brings test-time training to robot policies(6 posts)→
More from Embodied
- Amazon and Google sold 600M+ smart speakers, so why no AGI-era successor? — julianlehr · 2026-09-11
- ECCV26 Oral: Flow Matching Enables Single-Stage Multi-View Point Cloud Registration — ducha_aiki · 2026-09-11
- Polish developers build iPhone app that detects nearby Meta smart glasses — Low-Honeydew6483 · 2026-09-11
- Ant's Afu health AI hits 150M users, unveils AI+hardware health alliance at Bund Summit — APPSO · 2026-09-11
- Johns Hopkins Launches Full-Stack Hands-on Robot Learning Class with SO-101 Arm Kits — _krishna_murthy · 2026-09-11
- SyncWorld: In-Context Robot World Model Simulates Unseen Views and Embodiments Zero-Shot — ChongZzZhang · 2026-09-11