Fig. 3: Track used for testing the driver model.
From: Human-like driving behaviour emerges from a risk-based driver model
a The track contains four road and three traffic scenarios. The four road scenarios are (1) Curve radii: R100m, R200m, R300m and R400m, (2) Lane widths: 2.5, 3.0, 3.5 and 4.0 m, (3) Obstacle avoidance: A car was parked on a 3.5-m wide straight section such that 0.9 or 1.4 m of the car-width encroached on the road to simulate narrow (on) and wide (ow) obstacles, respectively. (4) Roadside furniture: A 200-m long row constituting of 10 cars was placed either outside the left lane boundary (asymmetric) or outside both lane boundaries (symmetric). The three traffic scenarios are (1) Car following: Two cars travelling at a constant speeds of 12.5 m s−1 (cfs) and 15 m s−1 (cff) along the lane centre on different straight sections were followed. (2) Overtaking: Two cars travelling at constant speed of 7.5 m s−1 (ovs) and 10 m s−1 (ovf) were overtaken using a 3.5 m overtake lane. (3) Oncoming traffic: Two cars travelling at a constant speed of 5 m s−1 on the 2-m wide oncoming lane, approached the ego car. The first oncoming car drove on the lane centre (onc). The second car was offset 0.3 towards the ego car. b To identify realistic values for the parameters of the DRF driver model, we replicated the track in a driving simulator and one volunteer drove 10 times ‘normally’ (blue) and 10 times in a ‘sporty manner’ (red). Speed and lateral deviation from the lane centre are plotted as a function of the distance travelled along the lane centre of the track. The speed and lateral deviation trajectories of the DRF driver model, for the most part, lie within the ±σ limits of the experimental trajectories. The ‘sport’ parameter setting consistently drives faster than the ‘normal’ setting and in both cases shows similar trends in acceleration braking as shown by the human. The driver model maintains itself within the lane boundaries, while exhibiting satisficing (i.e., not always following the lane centre), even in conditions that were not experienced during parameter estimation.
