Automatic driving model training method and data processing equipment

Abstract

The automatic driving model training method and data processing device provided by the present application obtain experience data through parallel exploration and learning by the second automatic driving model in multiple learning modules, and then store the filtered experience data to a shared target for playback The multiple second automatic driving models then replay the experience data in the cache according to the shared target and the network parameters of the first automatic driving model to update their own network parameters to continue exploring and learning. And when certain update conditions are met, the first automatic driving model is updated in combination with the network parameters of the second automatic driving models. In this way, during the whole training process, multiple parallel second automatic driving models are used for exploration and learning and experience data sharing, which improves the efficiency of training, and the first automatic driving model is updated by screening and controlling the shared experience data. The parameter update conditions of the model network parameters can make the whole learning process more stable.

Description

technical field [0001] The present application relates to the technical field of automatic driving, in particular, to an automatic driving model training method and data processing equipment. Background technique [0002] Automatic driving technology is one of the most practical applications of artificial intelligence in the real world. Large-scale implementation of automatic driving technology can effectively reduce the faults of human drivers, such as fatigue driving, distraction, speeding and dangerous driving, etc. Economic and social losses caused by unsafe behaviors. [0003] The realization of automatic driving technology currently mainly relies on the automatic driving model based on machine learning. The automatic driving model determines the action strategy based on the driving environment data collected during driving, so as to control the vehicle to automatically perform corresponding actions. In some training schemes for autonomous driving models, a virtual dri...

AI Technical Summary

Problems solved by technology

[0004] However, in the training schemes of these autonomous driving models, there may be a problem that the training speed is fast but the training process is unstable, or there is a problem that the training process is relatively stable but the training speed is slow, and it is difficult to balance the training speed and the stability of the training process

Images