Oussama Khatib visit

Yesterday Oussama Khatib from Stanford visited and give an great talk on human-friendly robotics, which is a theme he has been working on for a while. The talk had several different sub-topics, but the two key issues that resonated the most with me were that of how to assure robots can interact safely in a human environment, and how to uses robotics-based analysis to build better models of human beings.

In the context of safety, the idea is that robitic systems in human environments should not just include software that makes the behave in a safe way, but should have a hardware design that precludes dangerous behavior even if the saftware goes berserk. A trivial example of this concept might be to make a robot out of Jell-O, so that it could not possibly hurt you. On the other hand, making the robot out of very safe "stuff" is generally at odds with high performance, whre you might want hight speeds and good load carrying capacity. The challenge he addressed was to get both performance and assured safety, at least for some parts of the system, and he did this by (essentially) using a subtle analysis of the relationship between performance as a function of response time. Basically, strong robot muscles rarely need to move quickly.

We also had a great lunch and discussed the parallels between economic cycles driven by psychology, and similar cycles in the research community (related to how robotics was oversold back in the 1980's and didn't recover credibility until the late 1990's).