Escape and path point tracking methods of skid-steered mobile robots under undulating terrain conditions

• A novel recognition method for the SSMR's trapped state is proposed. • An escape trajectory model enables rapid path re-tracking after entrapment. • A hybrid reward function combines artificial and environmental incentives. • An escape strategy is developed based on the robot's internal state. • Comprehensive outdoor orchard experiments validate the proposed method. Skid-steered mobile robots (SSMRs) offer the advantages of a simple structure and good terrain adaptability, making them widely used in agricultural production. The ground of outdoor orchards is generally undulating and unpaved. When an SSMR operates on the undulating road, some tires lift off the ground and cannot provide driving force, thus bringing the SSMR into a trapped state due to insufficient driving force. Moreover, the undulating state of the outdoor orchard ground cannot be measured in real time due to the influence of vegetation coverage. Therefore, the SSMR cannot avoid the undulating areas that may cause some tires to lift off the ground through path planning methods. In this paper, the principles for the SSMR to be trapped due to partial tire lifting off the ground were investigated, and a trapped-state recognition method based on internal sensor data of the robots was established. On this basis, an escape action generation model of SSMR is designed, and the model training was completed based on deep reinforcement learning. Finally, the simulation experiment and physical experiment of SSMR escape are conducted. The experimental results showed that the methods proposed in this paper could effectively recognize the trapped state of SSMR and achieved a 90 % success rate of escape.