has announced patents related to the training of robot control models.

The QiChaCha APP displays that recently, Ubtech applied for a patent for "training methods, devices, electronic equipment, and storage media for robot control models." The patent abstract shows that the method includes: training and processing the initial motion control model based on the first total reward function to obtain the first motion control model; training and processing the initial navigation control model based on the second total reward function to obtain the first navigation control model; maintaining the parameters of the first navigation control model unchanged, training the first motion control model based on the first predicted speed output by the first navigation control model, the second sub-reward function, and the first total reward function to obtain the second motion control model; fine-tuning the parameters of the second motion control model and the first navigation control model to obtain a model for controlling the robot to avoid obstacles. Through this application, the accuracy and efficiency of robot motion control can be improved through reinforcement learning between the navigation control model and the motion control model.