Enhanced intelligent driver model for two-dimensional motion planning in mixed traffic

This study aims to model two-dimensional (lateral and longitudinal) motion of an Ego Vehicle (EV). Intelligent Driver Model (IDM) is enhanced for this purpose. All the surrounding obstacles (vehicles and road boundaries) are considered as stimuli that elicit driver's reactions (lateral and longitudinal accelerations). A Virtual LiDAR sensor on the EV perceives the surrounding. Proximity and speed are used to compute longitudinal and lateral components of average relative velocity. A new formulation is presented to compute the effective gap considering all the surrounding obstacles. It is composed of four components: front, rear, left, and right. The former two affect longitudinal acceleration, while the latter two influence lateral acceleration of the EV. A discretionary lane-change model is also developed to account for the driver's deliberate (not just mandatory) lateral movements. A total of 13 model parameters are used, of which eight are newly introduced. These model parameters are calibrated using 1050 human-driven car trajectories from Next Generation SIMulation dataset. Taguchi's fractional factorial design principle is used to optimize the parameters. Gray relational analysis indicated that the three newly introduced parameters to be the three most influential parameters. Validation using another 450 car trajectories resulted in a mean radial error of 3.97 m over a horizon of 10 s. The performance measures of two-dimensional motion planning (model error, longitudinal error, and mixed gap error) are found to be better than those reported in the literature. The developed human-like AV-IDM offers a proof of concept.