Slope path tracking control of agricultural wheel-legged robot based on virtual sensing radar and two-level deep neural network

The continuous development of smart agriculture puts forward the requirement of high accuracy slope path tracking for the agricultural wheel-legged robot. Compared to flat terrain, path tracking control on sloped terrain faces the obstacle of motion instability of the wheel-legged robot induced by the slope gravitational force component, which causes instantaneous steering center to offset. To address this problem, this study proposed a slope path tracking control algorithm by combining the methods of virtual sensing radar and two-level neural network. Firstly, the kinematic and dynamic models of the wheel-legged robot are deduced, from which the crucial factors affecting control accuracy of slope path tracking are recognized. Secondly, this study constructs the slope path tracking control algorithm, in which the virtual sensing radar is utilized to realize route perception, and the two-level neural network is employed to provide drive motors’ speeds to adapt to path tracking on different slopes. Furthermore, the corresponding compensation methods of the identified impacting factors are embedded in the proposed algorithm, including the lateral tracking deviation factor, heading angle deviation factor, slope change factor, and slip rate factor. Finally, the co-simulation model of slope path tracking control is constructed, including the multi-body dynamic model of the wheel-legged robot in RecurDyn and the proposed slope path tracking algorithm complied by Python. Subsequently, the simulation tests of the wheel-legged robot are carried out under various slope angles and velocities. The results reveal that the proposed algorithm’s effectiveness and accuracy are superior, with tracking errors reduced by more than 47.2% compared to an optimized pure pursuit algorithm.