Mo Zhihang launches the world's first cognitive model of autopilot DriveGPT,4 announces the latest development this month
CTOnews.com February 17 news, with the popularity of ChatGPT throughout the network, its Transformer model and "human feedback reinforcement learning (RLHF)" technology have once again attracted the attention of the industry.
Mo Zhixing announced today that the world's first self-driving cognitive model has been officially upgraded to DriveGPT, and its latest and important developments will be announced at the eighth HAOMO AI DAY in April 2023.
According to reports, in the field of autopilot, the Transformer model was first introduced into the data intelligence system MANA in China.
In January this year, Miao CEO Gu Weihao said that the self-supervised cognitive model of self-driving adopted ChatGPT ideas, adopted RLHF (human feedback reinforcement learning) technology, and continuously optimized the cognitive decision-making model of self-driving by introducing real human driving takeover data.
He said that with the help of this large model, in the generally recognized difficult scenarios such as turning around and around the island, the scene pass rate has increased by more than 30%.
Nomo said that DriveGPT has completed the model building and the first phase data running, and the parameter scale is comparable to that of GPT-2. Next, it will continue to introduce large-scale real takeover data to continuously improve the evaluation effect through the intensive learning of human driving data feedback. At the same time, DriveGPT will also be used as a cloud evaluation model to evaluate the driving effect of small car models.
CTOnews.com briefly introduced that the predecessor of Mo Zhixing was the Intelligent driving foresight Department of the Great Wall. In 2019, Mo Zhixing was hatched by Great Wall Motor and became an independent self-driving artificial intelligence technology company. At present, Great Wall Motor is not only a major shareholder but also a major customer of Maomo Zhixing.
Expand reading: "Mo Zhihang has released China's largest self-driving intelligence center MANA OASIS, with 6.7 billion floating-point operations per second."
