The modelling problem
Wheeled skid-steer machines generate motion by creating different wheel speeds on each side of the vehicle. During a turn, the tyres interact with the ground both longitudinally and laterally. A single isotropic traction coefficient therefore compresses two distinct physical behaviours into one value, limiting energy prediction and terrain interpretation.
This paper tests an anisotropic alternative that estimates separate longitudinal and lateral traction coefficients from onboard measurements. The aim is an interpretable model that can become part of a terrain-aware vehicle digital twin.
Test platform and data
The experiments used a four-ton CMTI MT1800B skid-steer mining platform on soft surface terrain. An approximately 18-minute run included different forward, reverse and turning manoeuvres. Sensor signals were filtered using a Kalman approach, while a Simscape-derived model estimated the changing normal forces at the wheels.
Parameters were identified over individual movement segments. Performance was assessed in two ways: how closely each model represented power within the calibrated segment, and how accurately those coefficients predicted energy use in the following segment. This second test is important because a model that only explains the data used to fit it has limited operational value.
Findings
The anisotropic formulation consistently produced a lower modelling error than the conventional isotropic formulation. It also improved forward energy prediction for nearly every movement category, with the clearest advantage during manoeuvres dominated by right-hand motion. The few exceptions were small for left-skid and reverse progressive-left movement.
The estimated longitudinal coefficient agreed well with an independent pull test, providing a physical cross-check that did not rely solely on optimisation error.
Why it matters
Separating traction directions gives a digital twin a more realistic description of tyre–terrain interaction. It could support route-energy estimation, mission planning, detection of changing ground conditions and safer autonomous control. It also keeps the terrain parameters physically interpretable, which helps engineers understand why the predicted demand changes.
Limits and next steps
The evidence comes from a limited surface run rather than a full range of underground terrains and loading states. More diverse, controlled tests are needed to establish how stable the coefficients remain across moisture, grade, payload and manoeuvre type. The next modelling step is a continuously updating, filter-based estimator that can converge during operation instead of fitting isolated segments after the event.