New AI Learned Parkour From Just 30 Seconds Of Video

Two Minute Papers

Summary

This video presents a fascinating new AI that can learn complex parkour moves from an incredibly small dataset, just 30 seconds of video. It addresses the common problem of AI either moving unrealistically or being too rigid, by combining imitation of human movement with goal-oriented adaptation. The AI uses a unique "two-classroom" training method and a "judge-athlete" system to continuously improve. While it achieves stunning results, successfully navigating challenging obstacle courses, the researchers acknowledge limitations such as lower success rates on very long levels and occasional unnatural recovery motions. The video emphasizes the importance of open-access research, highlighting that the paper and potentially the code are freely available.

Chapters

  1. 0:00Introduction to AI Parkour Problem
  2. 0:55Failures of Previous Techniques
  3. 2:19New AI Technique's Capabilities
  4. 4:14How the New AI Works
  5. 5:16Discriminator Training and Limitations
  6. 5:55Lambda AI for Research
  7. 6:16Free Research Paper Availability

Topics covered

  • New Hybrid AI Training Method
  • Learning Parkour from Minimal Data

+6 more topics (full content when you follow the channel)

Transcript excerpt

This is one of the most fun papers I've read in a while. You will love this one. So, what if you knew every athletic move perfectly, but had no idea when to use which? A perfect vault is useless if the table moves away. This is the strange problem of teaching parkour to a virtual human. Systems that copy us are beautiful, but brittle, repetitive. Systems that chase goals adapt, but often stop moving like us. But this new way I work says, "Enough is enough. Let's do both." It promises to remember exactly how humans move, yet is able to break the script when needed. Hmm. And get this, here's the crazy part, the training data. They say that all you have is only 30 seconds of internet parkour to pull this off. That, fellow scholars, sounds impossible. You see, previous techniques were not up to the task. This guy moves like a drunkard. Not to worry, try again. And >> [laughter] >> Yep, get s…