CoRL 2026 slide trains an LSTM in 1 minute to predict robot joint torque
shaohua0116 · x · 2026-11-11
Sharing a slide from a CoRL 2026 talk, the author jokes that 'probably most students don't know what LSTM is' anymore.
The slide shows step 2 of Neural External Torque Estimation (NEXT): using motor current × torque constant (K) = joint torque as supervision, an LSTM is trained on just 1 minute of data to predict free-space torque from joint states—teaching the robot to predict itself.
More from Fun
- After hours of AI brainstorming, founder concludes AGI hasn't arrived yet — burhop · 2026-11-11
- 7 years after her PM2.5 project at 13, she's now doing AI research at Stanford — alfcnz · 2026-11-11
- Anthropic engineer's Fable 5.2 render called AAA-level, but critics say humans are the real bottleneck in AI slop — teortaxesTex · 2026-11-11
- Meme: YAMLs, YAML Everywhere — AI dev workflows drowning in YAML — capetorch · 2026-11-11
- Grug comic nails AI hype: make sand think, get big big bananas — burny_tech · 2026-11-11
- Interactive 3D mixology with real-time fluid physics, built with Gemini 4 and WebGPU — hapticdata · 2026-11-11