Unmanned ground vehicles (UGV) have emerged from research institutes into the outside world over the last decade. For example, UGVs are already part of everyday operations in some mines and harbors around the world. Furthermore, self-driving cars have recently been widely discussed in the news and are expected to be on the market within the next five to ten years. Nonetheless, these UGVs are designed to operate in structured environments and cannot negotiate off-road terrain. Moreover, the reliability of state-of-the-art human detection systems is still not good enough to ensure safety at all times in some application areas.
This thesis proposes novel methods that address these limitations of current UGV systems. It presents the whole chain of developing a UGV system for off-road environments, but the main result of the thesis is to describe the novel terrain traversability analysis methods developed for unstructured environments. It also presents a new, efficient representation of traverasbility mapping and proposes two new approaches for traversability classification that exploit this representation. Furthermore, it presents two innovative methods for augmenting traversability with ultra-wideband (UWB) radar data. Since UWB radars can penetrate some amount of vegetation, the developed methods enable the clearance of obstacle-free vegetation (an area of vegetation that could be driven through) from the generated traversability maps, which is not possible with current state-of-the-art methods.
The thesis also presents two novel human detection methods that exploit 2D LIDAR (Light Detection and Ranging) and radar data, respectively. The LIDAR-based human detection method is computationally very efficient, whereas the radar-based method (utilizing only radar data) demonstrates the potential of radar sensors, typically more robust against adverse weather conditions, in human detection applications. The developed methods provide valuable insights into exploiting additional sensor modalities to supplement traditionally used, camera-based human detection methods.
The performance of all the developed methods was evaluated by conducting extensive field experiments using real UGV systems. The results demonstrate that the methods proposed in this thesis enable safe navigation performance for UGVs, even in densely vegetated, populated environments.
- , Supervisor
- , Advisor
- Jari Saarinen, Advisor
|Publication status||Published - 2017|
|MoE publication type||G5 Doctoral dissertation (article)|
- traversability estimation, obstacle detection, human detection, ultra-wideband radar, unmanned ground vehicle