Adaptive Load Control of Quadruped Robots Using Neural Networks

In the execution of tasks, quadruped robots frequently encounter scenarios necessitating load-bearing operations. To enhance the load-bearing capacity of quadruped robots in practical tasks, this paper proposes an adaptive load framework integrated with neural networks. Initially, a neural network estimator is devised, capable of estimating the mass and position of loads placed on the robot's back through data such as the torque of the robot's joint motors and the IMU of the body. The estimation results are then transformed into six-dimensional force applied to the robot's body. By modeling the dynamics of the quadruped robot, the external forces provided by the load are incorporated into the dynamic equations of the quadruped robot, achieving the effect of adaptive load. This method has been tested in both simulation and real-world environments, validating the effectiveness of the proposed approach.