Simultaneous Localization and Mapping

This project was done as part of a class (CSCI5980). The goal of the project was to guide a Pioneer - I robot about and indoor corridor environment and simultaneously map the environment. The robot was fit with a SICK LMS lidar sensor and ultrasonic rangefinders. The lidar sensor provides accurate range measurements with a 180 degree scan and a range of 8m. The accuracy of the lidar meant that the ultrasonic sensors were not needed.

The problem with dead-reckoning with mobile robots is that the odometry sensor information is not very accurate, and errors introduced to the system continue to grow unless checked. These errors are large enough that it cannot be ignored when creating a map from just the raw information. In order to reduce errors in odometry, a localization algorithm was developed that used the lidar sensor to find trackable features in the environment. The Extended Kalman Filter was used to incorporate these correction into the pose estimates to produce a better map.

The images below show maps of a corridor in which the robot traveled. The plots show that the errors in odometry become substantial as the robot moves along the environment with a growing pose uncertainty ellipse. When features in the environment are tracked, this error can be reduced, and when the robot reaches a landmark that it has seen previously, the uncertainty can be back-calculated to improve the map.

View the .fig file for this map to see the doorways and corners that are detected as trackable features.

Project Report