How Do Tracked Robots Achieve Millimetric Path Tracking?

In the world of tracked vehicles, like Mars rovers or rescue robots, staying on course is a tough job. A new study introduces a slip‑based predictive model, meaning it accounts for the tracks’ sliding on the ground. On hard surfaces, the vertical load concentrates under the drive wheels, allowing the vehicle to be treated almost like a four‑wheel car. Researchers built a nonlinear two‑track model that calculates track forces using a simplified shear‑displacement theory. This model feeds a Model Predictive Controller (MPC) that, unlike simpler kinematic models, anticipates the forces at play. Tests include constant‑speed cornering and spiral maneuvers, both in simulation and on a real five‑track prototype on asphalt and stabilized roads. Results show lateral deviations under 0.25 m and typically between 0.02‑0.08 m RMS. In practice, the vehicle follows the path almost perfectly while using less computing power. This approach promises more precise navigation for robots on firm terrain, enabling more complex and safer missions.

The slip‑based model relies on shear displacement theory, which describes ground deformation under tracks as a mix of sliding and compression. Practically, the longitudinal force F is computed as F = k·δ·exp(-c·δ), where k is ground stiffness, δ the shear displacement, and c a damping coefficient. The two‑track model treats left and right side forces separately, capturing differences in grip or load. The MPC uses a prediction horizon of 10 steps with a 0.05 s sampling interval, optimizing a cost function that penalizes both trajectory error and excessive steering commands. Local linearization of the nonlinear model occurs at each step, reducing computational load compared to a full‑nonlinear approach. Experiments show that, compared to the kinematic model, RMS error drops by about 70 % thanks to the slip‑based model’s ability to anticipate lateral force variations. Moreover, controller stability is ensured by a steering saturation constraint, preventing unrealistic command requests. This method suits vehicles on compact terrain, where ground response is predictable, but further validation is needed on softer or irregular surfaces, where the shear model should incorporate plastic deformation effects.