Tracking moving objects with a mobile robot

My main research interest lies in the area of probabilistic state estimation techniques, which enable a mobile robot to continously determine the current state of its environment.

One example is the use of particle filters for tracking the activities of people in the surrounding of a mobile robot. Please click on the image below to obtain a visualization of our robot Rhino tracking up to 6 persons in the corridor of our department.

You will see the real camera images and a VRML visualization of the robot's state estimates at almost the same point in time.

Particle Filters at Work

State estimates are based on features extracted from the data provided by the robot's laser range scanners. Features are tracked by particle filters, which represent the robot's belief of the persons' current positions. Like for example Kalman filters, a particle filter maintains a probability density over time. However, the probability distribution is represented as a set of samples (particles) drawn from it, and the density is updated using resampling techniques. In contrast to parametric methods like Kalman filters, the use of samples allow to represent arbitrary densities.

The following animation shows how the particle filters evolved over time, while the robot was tracking the people in the corridor. The colored dots show the individual samples of the sets.

State estimation is based on features extracted from laser scans. Local minima in the range profile indicate people. The range profile and the minima extracted are shown at the bottom of the images.