Cynthia Breazeal (MIT Media Lab’s Personal Robotics Group) gave a guest lecture at CMU’s Robotics Institute called “The Social Side of Personal Robots”. The main idea is that robots can’t be programmed to do everything, so they’ll need to be able to learn on the spot in order to be useful. Most people who will require robot assistance probably aren’t robot programmers, so robots will need to be able to learn through natural, human-centric social interaction. That’s proving quite a hurdle, as you’ll see in her video presentations. This leads to some interesting experiments where an instructor must communicate how to perform a task without speaking, which could then inform what non-verbal cues a robot should pay attention to in order to yield the best results from a learning interaction. There’s some great footage of the fluffy Leonardo (a robot co-developed with the late Stan Winston’s award-winning special effects studio) interacting with complete newbies (who responded to a Craig’s List ad).
Another interesting strategy they’re exploring are little computer games as simulations where human players interact with a robot (in this case, the dexterous mobile robot Nexi) online to complete a given task. The data can then be transferred to a real Nexi stationed at a museum, where it can interact with visitors to see how things might play out differently in the real world. The end goal is to teach robot behavior through crowdsourcing, but as you’ll see sometimes the robot may learn unwanted behaviors from their human instructors :)
The presentation is just over an hour long, but it’s definitely worth watching when you have time.
via [IEEE Automaton]