Tsinghua and partners report collective robot learning results across 96 real robots
从学习人类到学习机器人同伴，清华大学联合无问芯穹、智元提出具身智能规模化落地新路径
- Who
- Tsinghua University, Infinigence AI, and AgiBot
- What
- proposed and validated the Collective Experience Learning paradigm in a cluster of 96 real robots
- Quote
近日，清华大学、无问芯穹与智元联合提出群体经验学习（Collective Experience Learning，CEL）新范式，并在96台真实机器人组成的集群中完成规模化验证。
What happened
Tsinghua University and its partners proposed Collective Experience Learning (CEL), which lets robots learn from peers’ successful experiences, and validated it across 96 real robots. Two robots excluded from training achieved 97.5% overall success across 600 tests covering five tasks. The approach combines a fault-tolerant asynchronous training system built on RLinf with failure-guided experience retrieval and fine-tuning to target improvements using past successful operations.
Read next
See the full storyTsinghua, Infinigence AI, and AgiBot Test Collective Experience Learning Across 96 RobotsThe story so far
Tsinghua University, Infinigence AI, and AgiBot proposed Collective Experience Learning (CEL) and tested it across 96 real robots; according to Qbit, two robots excluded from training achieved 97.5% overall success across 600 tests covering five tasks. CEL lets robots learn from other robots’ successful operations. It combines a fault-tolerant asynchronous training system built on RLinf with failure-guided experience retrieval and fine-tuning, using past successes to target improvements where robots fail.
Keep up with AI, one email a day
Model releases, price changes and compute developments, with original sources. Free, confirmed by email, and easy to unsubscribe. Read a past issue · 2026-10-30