Date: 04/02/09

Speaker: M. Greytak

Title: Planning to Learn:  Robust Motion Planning for Mobile Robots with
Autonomous Learning Strategies

Abstract:
To generate robust motion plans for mobile vehicles, we need to be
able to predict the probability that the vehicle will collide with
obstacles when following the plan.  Two factors that affect the
collision probability are environmental disturbances and modeling
error.  We have developed methods to predict the collision
probabilities analytically without the use of costly Monte Carlo
simulations.  Furthermore, we can predict the convergence of the model
parameters during online learning, thereby closing the loop between
learning and planning.  Motion planning with learning predictions
leads to some interesting emergent learning strategies.