Month: March 2014

Making robots behave

I normally write about things that are somehow related to what I do, but last week I attended to a seminar that I really enjoyed, and I think it is worth making a short summary here. The title was “Making Robots Behave”, and it was presented by Leslie Pack, (MIT).
As you may have already guessed, the seminar was about Artificial Intelligence and robot behavior. In particular, the problem they wanted to address is how to solve robot uncertainty. The robot has a state estimation from which it creates its beliefs and uses the result to pick up the next action. Then a planner takes the action into consideration and tells the robot to execute the next planned action. From the result of the action a new input is generated and the robot re-estimates the next state, initiating the cycle again. A simple feedback loop! (For more details I recommend you to have a look at the talk, available online).
One of the cool things about the demo we saw is that most of the time you didn’t knew how the robot was going to act next. It depended on the reads the sensor gave to it in each moment, plus the planning algorithm and the feedback from its previous actions (i.e., its own knowledge of the world surrounding it). Sometimes the robot even moved its arms away because they were in the middle of its field of view.
I enjoy this kind of things because it reminds me somehow to a science fiction tale (the last question by Isaac Asimov) where humans ask a computer how to lower the entropy of the universe. After thousands of years, when the computer gets the response, it ends up knowing how to answer questions like the meaning of life. In the robot’s particular scenario the universe is the room, where it enters knowing nothing. When it exits, it is able to recognize the environment and the different objects found in its way. Whether the robot can learn from this experience and teach the actions to take in similar scenarios is something that we will have to wait to see.