Using Sigma Point Kalman Filters for Robot State Estimation for Robocup 2004
by
Griff Hazen
Each year the Robocup Competition attracts universities from across the world to the challenge of creating teams of autonomous robots for competing in soccer. Our department competes in this competition be developing teams of AIBO robots. For last year's competition, our team used a combination of a Particle Filter and Kalman Filter to estimate robot position and ball position independently. Because of the interdependence between robot position and ball position, it may be advantageous to combine the two estimates into the same state space. This may allow our system to be more efficient and robust by using an estimator from the family of Sigma Point Kalman Filters to estimate the combined 7 dimensional state space. Sigma Point Kalman Filters have another very important advantage over traditional Extended Kalman Filters in that they help better estimate nonlinear observations and processes, both of which are prevalent in our state space. I will present various experiments performed on our AIBO robots comparing the various approaches to state estimation.
Advised by Dieter Fox
EE1 037
Wednesday
May 26, 2004
3:30 - 4:20 pm