Autonomous Driving is one of the main subjects of academic research and one important trend in the automotive industry. With the advent of self-driving vehicles, the interest around trajectory planning raises, in particular when a customer-oriented analysis is performed, since more and more the carmakers will have to pay attention to the handling comfort.
With that in mind, an experimental approach is proposed to assess the main characteristics of human driving and gain knowledge to enhance quality of autonomous vehicles. Focusing on overtaking maneuvers in a highway environment, four comfort indicators are proposed aiming to capture the key aspects of the chosen paths of a heterogeneous cohort.
The analysis of the distribution of these indicators (peak to peak lateral acceleration, RMS lateral acceleration, Smoothness and Jerk) allowed the definition of a human drive profile. These characteristics were then transferred to the simulation environment to create a pseudo-natural trajectory planning strategy, via polynomial fitting and spline optimization. This strategy differs from the standard approach of trajectory planning, where absolute minimums of cost functions are pursued.
The polynomial and spline fitting techniques reached satisfactory results and are evaluated as valid procedures to imitate a natural human behavior in a simulation environment (also applicable to control the trajectory of AD systems) and raise a question about whether a human-like behavior can be subjectively perceived as better driving, despite not presenting optimized comfort indicators.