The purpose of this study is to define the startup slip phenomenon of the previously designed wheeled-climbing robot and to estimate the slip distance, thereby compensating for the robot’s positioning error.
First, a dynamic model of the robot’s startup slip process was established, and the robot’s startup state was calculated by monitoring the driving current. Subsequently, a slip estimation algorithm based on dynamically extending Q-learning (DEQ) was proposed, and several estimation strategies were tried for different states. In addition, a faux-Gaussian reward function was designed to provide feedback to the strategies.
The experimental results show that the DEQ method performs better than the theoretical calculation method. Average positioning error of the robot at startup was reduced from 0.96% to 0.05%.
Compared with other methods, the DEQ method requires less sensor data to be monitored, does not need to calculate or simulate the complete process of slip, and does not require other positioning sensors.
