The LEGO Mindstorm NXT robotics system is an excellent testbed for research in machine learning and artificial intelligence. At Knuthlab Robotics at the University at Albany, we are developing intelligent instruments using LEGOs.

Our first instrument is a robotic arm that is designed to locate a characterize a white circle on a black background using the LEGO light sensor. It relies on Bayesian inference, which is implemented using a technique called Nested Sampling, which was developed by John Skilling. This software allows the robot to learn the characteristics of the circle using the light sensor data that it has collected. The real advance here is the inquiry engine, which uses Bayesian adaptive exploration to decide which measurements to take next. It does this by considering all the possible measurements that it could take, and computes the expected gain in information from each possible measurement. It then chooses to take the measurement with the greatest expected information gain. The process then repeats as the robot learns about the circle.

The system is easily generalized to solving other problems, such as exploring rooms, interpreting people’s emotions, and doing real science.

We recently presented our research at the MaxEnt 2007 workshop in Saratoga Springs NY. Below are links to a video of the talk, my slides, and our research paper.

Video: Designing Intelligent Instruments, K.H. Knuth

Slides: Designing Intelligent Instruments, K.H. Knuth

Research Paper:

Knuth K.H., Erner P.M., Frasso S. 2007. Designing intelligent instruments. K.H. Knuth, A. Caticha, J.L. Center, A. Giffin, C.C. Rodriguez (eds.), Bayesian Inference and Maximum Entropy Methods in Science and Engineering, Saratoga Springs, NY, USA, 2007, AIP Conference Proceedings 954, American Institute of Physics, Melville NY, In press.