Probabilistic Geophysical Modelling and Learning Navigational Maps for Social Robots
Simon O'Callaghan
NICTA SML SEMINARDATE: 2013-08-01
TIME: 11:15:00 - 12:15:00
LOCATION: NICTA - 7 London Circuit
CONTACT: JavaScript must be enabled to display this email address.
ABSTRACT:
The talk will be split into two parts. Initially, I will present recent developments in the Geothermal Data Fusion Project. In particular, I intend to discuss results from test scenarios where information from multiple geophysical sensors were fused into a single probabilistic inversion to model the underlying geology. Early work on methods for communicating the high dimensional posterior to the geologists will also be presented along with an overview of the project's future direction. The second part of the talk will change gears and look at learning social trajectories for robots operating in environments shared with people. Observing human motion patterns is informative for autonomous path planning when the shortest path is not always the most optimal solution. The core ML algorithm employed here is the Gaussian process however I will also show some more recent work in which a Kernel Bayes' Rule approach is adopted to overcome some of the limiting assumptions made by the GP.
BIO:
