EmbodiedEvo tops RoboDojo at 38.95% success by letting robots inherit failed practice
thetripathi58 · x · 2026-10-15
Navers Lab and Einsia AI released EmbodiedEvo, a self-evolving coding agent for physical intelligence: one Playground produces reusable Skills & Lessons that transfer to unseen layouts without model-weight updates, scoring 45.08 average and 38.95% success—#1 on the RoboDojo leaderboard.
Key findings:
- Failures leave value behind: an agent never finished the conveyor task in practice but derived a working grasp-and-lift; a fresh agent reusing it went from 0% to 40% success across five new scenes
- Experience boosts execution: Astra/xhigh success on matched scenes rose from 8.50% to 30.17% with task descriptions alone
- Knowledge transfers across agents (+8.86 / +5.50 pp), persisting when both agents also get the wiki
- Especially strong on the Memory metric (71.90)
Paper, project page, and GitHub are public.
Related event: EmbodiedEvo: Coding Agent That Evolves Through Physical-World Experience(2 posts)→
More from Embodied
- SEAR: first large-scale controlled study of self-evolving robot agents, 7 LLMs, 330 tasks — xiye_nlp · 2026-10-16
- Podcast: subsea robotics startup Ulysses on operating 1,500m underwater and ocean carbon — blakemenezes · 2026-10-16
- Sony unveils $5,900 gold-plated Walkman with AI music enhancement — Polymarket · 2026-10-16
- ROBIB Meetup #3 returns to Ghent on Nov 19 with Physical AI talks, demos and panel — MarwaEldiwiny · 2026-10-16
- VLX-Go: Multi-Stage Vision-Language Training Boosts Robot Person-Following Among Distractors — omlab · 2026-10-16
- Ledger Atom Debuts as Agent-Approval Security Device; Critic Says a $100 iPhone 8 Beats It — adamamcbride · 2026-10-16