Podcast: Play in new window | Download
Subscribe: iTunes | Android | RSS
Jason talks with Peter Stone, director of the Learning Agents Research Group at UT Austin, whose team UT Austin Villa won both the standard platform league and the 3D simulation league of the RoboCup 2012 competition. They discuss why robot soccer is a good motivating application domain for machine learning, how the RoboCup competition got started and the kinds of teams that participate, the offshoot competition RoboCup Jr, why the machine learning technique known as reinforcement learning works so well in complex, dynamic environments, the role played by game theory in multiagents systems, Peter’s involvement in developing autonomous driving vehicles and what it’s like to run an artificial intelligence laboratory.
I’d love to get more shows on AI-related topics as I enjoyed the interview.
A couple bits I found interesting:
– the blending of various disciplines from mechanics, robotics, machine vision, strategy, etc.
– adapting the strategy algorithms to match the available sensory processing speed
– the counter example about the brute-force tweak-test-adjust-repeat cycle used by a Chinese researcher to get the right kind of leg movement
While listening I kept having flashbacks of the Sarah Connor Chronicles! 🙂
@Philippe Monnet – I like the AI stuff as well, so I’ll see what I can do. 😉