干旱区生态环境知识资源中心

This contribution reports on the unmanned and entirely autonomous land vehicle ION that traversed 47 km through the Mojave desert in the final race of the Grand Challenge 2005. After outlining the overall system architecture the article focusses on video-based path planning. A probabilistic model is introduced incorporating prior expectations on the path to be traveled. The observation model exploits the diversity of image features, namely disparity, texture, and local orientation. A Bayesian estimator spans an active search tree, that serves to continuously compute an optimal path, whose length increases with computation time. Results from real imagery of the Mojave desert indicate a high level of robustness of the proposed system.