A Novel Safety-enabled Belief-based Behavioral Decision-making Method for Autonomous Driving

The behavioral decision is one of the challenging tasks in autonomous driving because multi-sensors data for decision making may be uncertain and in high conflict. In this case, how to fuse the information to achieve a reasonable and cautious decision is an important issue. In this paper, we propose a novel belief-based behavioral decision-making method aiming to fuse the data from multi-sensors at the decision level to obtain a more cautious result. The data from multi-sensors is mapped to corresponding behavioral decisions which are taken as multiple pieces of evidence. Multiple pieces of evidence can be divided into several groups according to the most belief class. As the new evidence by combining the pieces of evidence in the same group is in conflict, a new belief-based biased combination rule is proposed to solve the conflict evidence fusion problem. On this basis, uncertainty is represented by committing the result to the proper meta-classes. The conflict and uncertain behavior decision will be reasonably committed to some proper results. The effectiveness of the proposed method has been tested and compared with other combination rules through a series of experiments.