N₀-TWAM gets project page, paper, code and caveats for its 46.3% robot result
imjustnewatai · x · 2026-07-29
This follow-up post provides the primary sources for N₀-TWAM: the project page, technical report, full HTML paper, and code repository.
It also clarifies a few important caveats:
- 46.3% is the macro average across 8 real-robot tasks, with 20 trials per task
- 30.0% is the strongest VLA baseline on the same suite
- Individual task results have large statistical uncertainty because the trial count is small
- The findings are author-reported and have not yet been independently replicated
The author says the “touch as the verifier for physical AI” framing and the idea of a tactile-data moat are their own interpretation, not claims from the paper itself.
Related event: N₀-TWAM Enables Robots to Predict Tactile Feedback(2 posts)→
More from Embodied
- Vinci2 Agent Outperforms GPT-5-mini in Proactive Assistance Benchmark — jiqizhixin · 2026-08-24
- Building U-BOT: Refining front chassis for battery and casters — _Stocko_ · 2026-08-24
- Matic robot vacuum mapping demo visualizes learning home layout — tristanbob · 2026-08-24
- NORINCO showcases explosion-proof humanoid robot Fuxi with real-time teleoperation — CyberRobooo · 2026-08-24
- Video Gen Consumes 70% of AI Tokens in China, Diverging from US LLM Focus — AccBalanced · 2026-08-24
- Retriever: A Framework for Asynchronous, Closed-Loop Robot Agents — ZeYanjie · 2026-08-24