I was at ICDL 2008 two weekends ago. This was my first time attending a conference, and so I can’t say whether it was good or bad relative to other conferences in my area (developmental robotics). I had a great time though, especially considering the rustic Pacific Coast scenery at Ansilomar.
A couple of highlights:
- I-POMDP: An Infomax Model of Eye Movement by Nicholas Butko, Javier Movellan — As far as the talks go, this was my favorite. It included an entirely intuitive though technically sound explanation of a clever application of a POMDP model to eye-tracking.
- Sensorimotor Abstraction Selection for Efficient, Autonomous Robot Skill Acquisition by George Konidaris, Andrew Barto — Again a very clear exposition of a particular method of applying RL ideas to robotics specifically.
- Internal State Predictability as an Evolutionary Precursor of Self-Awareness and Agency by Jaerock Kwon, Yoonsuck Choe — I meet Yoonsuck for the first time at the conference, though I’m very familiar with his work, both from his days at UT and now at A&M. One of the few people to articulate some truly fundamental problems in development, like how we come to understand that a pattern of neural activations corresponds to anything “out there” at all. His theory (which I share) is that a lot of the meanings of neural activations as a result of sensing are grounded in particular motor policies.
- Detecting the Functional Similarities Between Tools Using a Hierarchical Representation of Outcomes – Jivko Sinapov, Alexander Stoytchev — Sinapov gave a very smart and convincing presentation on how robots can come to understand tool use. The end goal of this research is to develop robots cabable of this kind of tool use reasoning.
I’ll have some more highlights later.
Entries (RSS)