Replacing hand-tuned RL rewards with a decision model to teach humanoids to walk
tw_killian · x · 2026-11-07
- Researcher Agneya Tharun trained two humanoids to walk from scratch as a controlled experiment: one with 12 hand-tuned rewards (stay upright, avoid jerky motion), the other judged by decision model elephant-1 on how naturally it walked.
- The experiment asks whether a decision model can replace the reward function in RL; results are in the linked video. The outer post itself is a joke meme framing.
More from Embodied
- Figure humanoid robot strolls through company HQ, nearly indistinguishable from a human — Distinct-Question-16 · 2026-11-07
- CoRL robotics conference sells out with hundreds on the waitlist ahead of Austin edition — tomaszbednarz · 2026-11-07
- Open-source AI wearable ring museo opens pre-orders at $144, works with any agent — bihanmahadewa · 2026-11-07
- InstantHMR open-sourced: real-time 3D human mesh recovery in 5.5ms per person on phone NPU — PurzBeats · 2026-11-07
- Muse starts rolling out to Meta Glasses, targeting hands-free AI agent control — armand_ruiz · 2026-11-07
- "2050 arrived early": a wave of humanoid robots on display — CyberRobooo · 2026-11-07