Multi-Unit State Estimation

Licentiate thesis, 2026

Any choice needs information, and good choices need good information. That holds for picking a path on a misty mountain, for deciding whether to trust the weather forecast, and no less for a vehicle. For safety-critical decisions, the information must be not only accurate but consistent: its reported uncertainty must match its actual error. For an articulated heavy vehicle, part of this information is the motion state. Stability-critical states like the coupling force, the articulation angle, and the lateral velocities of both units are not measured. These are the quantities this thesis estimates. Estimators for articulated vehicles have approximated the coupling: they linearize, assume small angles, prescribe inputs, or relax it by introducing a state. Paper A takes this last approach, a damper formulation that carries the coupling force as a filter state. Paper B keeps the exact coupling constraint and recovers the coupling force with its uncertainty. Uncertain parameters are carried as consider parameters, fleet-level knowledge in place of per-vehicle tuning. In large-scale simulation, the estimator of Paper B reaches median errors below one centimeter per second in lateral velocity, hundredths of a degree in articulation angle, and a few hundred newtons in coupling force. It is consistent at the median there, but over-confident on the tractor lateral velocity in the single test-track case. It runs in real time for up to four units, though only two are evaluated for accuracy. The main limitation is the static vertical-load assumption, under which accuracy is lost when the load shifts almost entirely to one side. This work shows that the exact coupling constraint is a viable basis for real-time, consistent estimation. The vehicle then knows its footing in the mist, and how far to trust it.