World Representation Using Terrain Maps: Enabling High-Speed Navigation

摘要

This report introduces a new terrain map technique for representing unstructured environments. Using an inexpensive nodding 2-D laser range finder this technique captures multiple terrain data sets and optimally fuses the data into a wrappable terrain map. Traditional terrain map implementations scan a terrain patch only once, thus integrate a single data point into the map. The nodding laser, allowing multiple scans of a given terrain patch, must fuse multiple data sets into the terrain map. A moving vehicle, with a limited pose accuracy, demands a data fusion technique that minimizes the effects of pose errors. This research investigated and tested a variance weighted statistical technique to optimally fuse terrain data sets into a terrain map. Simulations and experiments were conducted, which demonstrated the performance of the variance weighted technique as superior to classical statistical methods. DRDC implemented a variance weighted terrain map under the Miro architecture and tested its performance during the Autonomous Land Systems (ALS) demonstration. This terrain map technique performed admirably and its accurate terrain representation allowed the Raptor Unmanned Ground Vehicle to successfully navigate while avoiding obstacles and hazards.