A preprint testing a two-part driver model for road-truck energy-consumption simulations found that curve radius and driver differences stood out in speed analyses, while curve direction was weak by comparison. The model’s operational settings also varied across drivers and events, and uphill estimates often approached imposed limits.
A model with two jobs
The model divides driving into two linked tasks: a tactical part that generates a desired speed from environmental inputs, and an operational part that tracks it. The study asks whether this structure can reproduce realistic along-the-road driving behavior well enough for energy-consumption simulations.
The evidence came from two settings. Eight naïve professional truck drivers completed simulated drives of about 10 minutes on a country road. A separate field database supplied by Volvo Trucks covered 33 heavy-truck vehicles and about 1,000 logfiles from the OCEAN project.
Researchers estimated the operational-model parameters with box constraints and multiple starting points, using MATLAB’s standard fmincon interior-point method.
Curves tell a mixed story
One test examined how speed changed with curve radius. The simulator’s curve-speed exponent—a number describing that relationship—was 0.23, with a 95% confidence interval from 0.18 to 0.28. The paper reports that the tested value of 0.5 was rejected because it fell outside that interval.
At a reference curve radius of 250 metres, the simulator model using the estimated exponent gave a reference speed of 60 km/h. Driver variation was 16%, while the residual standard deviation—the remaining spread after the model’s terms—was 4.8%. Fixing the exponent at 0.5 gave a reference speed of 66 km/h and a residual standard deviation of 9.6%.
The field logs produced an exponent of 0.46, with a 95% confidence interval from 0.458 to 0.462. The paper describes that estimate as close to 0.5 but still reports 0.5 as outside the interval. Because field speeds plateaued for radii above about 500 metres, the analysis used curves from 20 to 500 metres; the paper notes that very small-radius curves could represent parking.
At the same 250-metre reference radius, the field reference speed was 54.5 km/h. Driver variation was 5.6%, the curve-direction effect was plus or minus 0.2%, and residual standard deviation was 17.7%—larger than the simulator’s 4.8% under the estimated-exponent model.
The statistical tests reflected that contrast. In the simulator, radius and driver showed very strong effects, while curve direction was not statistically significant (p=0.768). In the field data, radius and driver again showed very strong effects, but direction was only barely statistically significant (p=0.0123).
Where the framework strains
A separate speed-change analysis found that reported deceleration magnitudes were larger than acceleration values in the corresponding categories. The median was -1.16 for a 70-to-40 transition versus 0.52 for 40-to-70, and -1.06 for 70-to-60 versus 0.44 for 60-to-70.
Operational estimates differed across drivers and events. Downhill estimates generally stayed within the expected range, whereas uphill estimates often approached the imposed bounds, making uphill behavior harder for the model to capture.
The authors conclude that the model captures general driver-behavior trends relevant to energy-consumption simulation, but needs extensions for driver variability and environmental conditions.
The document carries an arXiv version 1 marker dated 20 Aug. 2026. The authors report financial support from the Swedish Energy Agency and acknowledge the OCEAN project and the FFI program.
Paper data and sources
Original title: Validation of a driver model for energy consumption simulations of road vehicles
Authors: Luigi Romano, Michele Godio, Fredrik Bruzelius, Pär Johannesson
Journal/Repository: arXiv
Status: Preprint, not yet peer-reviewed
First online: 2026-08-20
DOI: Not available
Original paper · Full text