Cao Ting says embodied AI is shifting from offline data training to self-evolution
Overview
Tsinghua AIR professor Cao Ting tells Leiphone that robots trained offline still fail in real settings, so research teams are turning to in-context learning.
She says her group's Zeva and Zeva-Ego models extract action-effect pairs from video, and that Zeva-Ego was trained on about 10K hours of unlabeled ego data.
Cao says factory deployment still relies on fine-tuning to reach 99% precision. She argues that physical self-improvement is needed for robots to learn beyond their training data.
Written by AI from the articles below · updated Nov 12, 4:17 AM ET
Check the sources:
Article timeline
The articles in this story. Times are ET.
- LeiphoneNewsCao Ting says embodied AI is now chasing self-evolution as data factories cool
AITsinghua AIR professor Cao Ting tells Leiphone that robots trained offline still fail in real settings, so teams are turning to in-context learning. She says her group's Zeva and Zeva-Ego models extract action-effect pairs from video, and that factory deployment still relies on fine-tuning for 99% precision. Zeva-Ego was trained on about 10K hours of unlabeled ego data, and she argues physical self-improvement is needed for robots to learn beyond their training data.
Heat Trend
- Current comparable heat
- 9
- Peak comparable heat
- 9Nov 12, 5:00 AM ET
- Change in past 24 hours
- –
The trend compares only continuously observed participants, so its scope may be smaller than the current heat total. Hover or tap for hourly heat; use arrow keys with a keyboard.