Autonomous Driving Considering Recommended Traversable Areas in Pedestrian Environment with Indistinct Road Boundary

Authors:

Abstract
The objective of this research is to develop autonomous mobility adaptable to the pedestrian environment with indistinct road boundaries using a recommended traversable area detection system. In previous research, we proposed a learning method for weakly-supervised semantic segmentation, which was effective in improving the performance of detecting recommended traversable areas. In this paper, we developed an autonomous navigation system based on a driving recommendation map using a recommended traversable area detection system. By setting appropriate weights to consider recommended traversable areas, generating paths in the direction of right and left turns while suppressing path generation to non-recommended areas was achieved. Also, by adding a new class of driving trajectory and sidewalk area, the system can selectively enter sidewalks in scenes where physical features and edges cannot be taken. Finally, the usefulness of the developed system in a real environment with indistinct road boundaries was verified.