Gear-Shifting Strategy Learning for Unmanned Tracked Vehicles Driven by Data and Knowledge Driven

The right gear-shifting timing is key to precisely longitudinal speed tracking of tracked vehicles. However, gear-shifting strategies often depend on comprehensive analysis of transmission systems and operating conditions, with many influencing factors and high modeling difficulty. In this paper an unmanned tracked vehicle gear-shifting strategy learning method via integrating driver data with prior gear-shifting knowledge was proposed. Gear-shifting operations of skilled drivers under various conditions were collected and analyzed for extracting driving behavioral characteristics. The Gaussian Mixture Model was used to cluster driver gear-shifting behavior, and gear-shifting strategies were designed based on gear-shifting point distribution features. Experiments show the proposed hybrid gear-shifting strategy model can considerably extract driver gear-shifting features effectively, forms excellent driving strategies similar to those of skilled drivers, and improve gear-shifting quality of unmanned tracked vehicles.