Abstract: On-line identifying soil parameters should address the multiple-solution problem caused by the coupling of soil parameters and the accuracy problem caused by state noises and measurement noises. For the multiple-solution problem, by taking advantage of the additional degree of control freedom of air-cushion-typed off-road robots for vertical forces, the g-EKF algorithm, logarithmic-mean algorithm and least-squares method are designed to identify, step by step, the three tractive forces-related soil parameters, the three vertical forces-related soil parameters and the two bulldozing resistances-related soil parameters. For the accuracy problem, several measures are implemented within the g-EKF algorithm in step 1 to reduce the influences of the state noises and measurement noises, consisting of selecting reasonable sampling points and rearranging and adjusting measurements; as well, in the logarithmic-mean algorithm in step 2, the logarithmic transformation is used to inhibit the impact of asymmetrically distributed noises on the mean calculation. Comparative experiments are conducted to examine the feasibility and accuracy of the designed estimation algorithms in each step, the propagated impacts of estimation accuracy of the forward step(s) on that of the later step(s), and the impact of estimation accuracy of the soil parameters on energy consumption optimization. The results show that the proposed step-by-step identification strategy and the corresponding algorithms are competent to solve the multiple-solution problem and the accuracy problem in the examined working conditions with different levels of noises

Key words: Air-cushion vehicle, Estimation error, Off-road robot, Parameter identification, Soil parameter