Design choices

• known geometry of the field
• predefined color specification of objects
• dynamic environment with possible crashes
• environment cannot be modified and global sensors are not allowed.

Range sensor: vision system (color based object recognition, detecting boards and lines, no intereferences with other sensors).

Localization method: map matching between sensor data and reference map.

A critical element for several map matching methods is noise given by isolated points and by occlusions of lines.

This noise is typical in crowded office-like environments and in the RoboCup environment.

We devise a new localization method that exploits the properties of the Hough Transform for robust and efficient map matching [Iocchi-Nardi, CSCC'99], [Iocchi-Nardi, RoboCup'99].
 
 
 

Hough Transform based localization

Hough Transform is a robust method for detecting lines from a set of points.

Hough demo
 
 

Map matching is performed in the Hough domain.

The displacement between the estimated pose of the robot and the real one is easily computed in the Hough domain.

Model of the environment: 7 segments

Application of the method

1. Small positioning error assumption (motor encoders)
2. In case of ambiguities:
   • Accumulation of more range data
   • Integration with absolute positioning methods (vision, compass)
3. Never during critical tasks (vision)

Performance

• robustness to noise given by isolated points and line occlusions
• efficiency O(n+m) = a few ms.
• accuracy depends on the discretization of the Hough plane (5 degree, 10 cm).
• precision depends on the precision of the range sensor.

Examples

Go to Control system