With the ability of vertical take-off and landing, the task path of an air-ground vehicle will be significantly shortened. Accordingly, the energy consumption will be greatly reduced. Through reasonable planning of the path, such vehicle can meet the high-efficiency needs of unmanned tasks and alleviate the global energy shortage problem. To design an optimal feasible path, this paper proposes a smooth path planning learning strategy considering mode switching. A new reward function of the Q-learning algorithm is presented, considering the influence of flight obstacle crossing parameters. To avoid the redundant flight distance and energy consumption caused by frequent high flights, the flight height correction is made in the update rule. Besides that, a path smoothing modification, called double yaw correction, reduces turning points and improves the path smoothness. It further reduces the energy consumption caused by the tortuous path. This modification also points out the direction of iterative learning and accelerates the algorithm convergence speed. Finally, the proposed strategy is verified on a 40m*40m map with 0-10m obstacle height. Results show that, the proposed strategy is effective to shorten 4.08m distance and plays the role of smoothing the path. Its convergence speed is faster than the traditional algorithm.
Keywords air-ground vehicle, path planning, mode switching, path smoothness, Q reinforcement learning