Large-scale Robotic Mapping using Gaussian Processes

Soohwan Kim (ANU)

COMPUTER VISION AND ROBOTICS SERIES PhD monitoring

DATE: 2012-03-29
TIME: 14:00:00 - 14:30:00
LOCATION: RSISE Seminar Room, ground floor, building 115, cnr. North and Daley Roads, ANU
CONTACT: JavaScript must be enabled to display this email address.

ABSTRACT:
Map building is a fundamental problem in robotics since maps are the robot's representation of the world and actions are made based on them. Recently, some robotics researchers have applied Gaussian processes to map building which is well-known as Kriging in the geostatistics field for interpolating sampled data. However, one critical issue of kernel methods including Gaussian process is the computational complexity of O(n^3) where n is the number of data points. Thus, it is impractical for large datasets which is common in outdoor robots. As part of my first-year thesis proposal review, I will provide a probabilistic and scalable mapping framework utilising Gaussian and Dirichlet processes, with some promising results on the simulated dataset. Future research directions will be discussed focusing on 1) approximation for large dataset 2) handling input noise and 3) demonstration with real data.