Human-like behavior generation for intelligent vehicles in urban environment based on a hybrid potential map

In this paper, we present a human-like behavior generation approach for implementing the autonomous driving functions along a predefined route in normal urban streets. Firstly, the surrounding vehicles are detected and classified into different categories based on their states of operation. Subsequently, a hybrid potential map, which includes a trajectory-induction potential and a risk-prevention potential, is constructed for each of the vehicles according to their categories. Finally, all of the potentials are integrated and used to generate an appropriate behavior for the ego intelligent vehicle. Particularly, the trajectory-induction potential, accounting for the apparent risks of the environment, is learned from real human-driven vehicle trajectories of the same situations in the daily urban traffic. The risk-prevention potential, accounting for the underlying risks of the environment, is constructed with multiple expert human drivers' instructions based on their knowledge and experiences. Therefore, a safe as well as reasonable behavior generation, which is close to a human-like performance, can be expected based on the proposed hybrid potential map. Experimental results in various typical but challenging urban traffic scenes have substantiated the effectiveness of the proposed system.