On February 17, Mayto Intelligence announced the official upgrade of its autonomous driving cognitive model to DriveGPT and will reveal its latest significant progress at HAOMO AI DAY in April 2023.
With ChatGPT going viral online, the Transformer model it adopted and the “Human Feedback Reinforcement Learning (RLHF)” technology it used once again drew industry attention.
In the field of autonomous driving, Mayto was one of the first companies to introduce the Transformer model into its data intelligence system MANA.
At HAOMO AI DAY in January this year, Mayto Intelligence CEO Gu Weihao stated that the company’s self-driving monitoring cognitive model, which the company launched, drew inspiration from ChatGPT’s implementation approach and used RLHF (Human Feedback Reinforcement Learning) technology to continuously optimize its autonomous driving cognitive decision-making model by introducing real human driving supervision data.
Mayto’s cognitive decision-making algorithm has gone through three stages of evolution. The first stage introduced end-to-end learning of individual scenarios, directly fitting human driving behavior. The second stage introduced massive normal human driving data through the cognitive model and achieved controllable and interpretable cognitive decision-making through prompts. The third and current stage is introducing real takeover data and beginning to try to use RLHF algorithm to learn from human driving takeover data in the large model.
Today, Mayto Intelligence officially announced the upgrade of its self-driving monitoring cognitive model to DriveGPT. Currently, Mayto’s DriveGPT has completed model construction and the first stage of data running. Next, DriveGPT will continue to introduce large-scale real takeover data to continuously improve its evaluation effect through reinforcement learning with human driving data feedback. Mayto’s latest important progress with DriveGPT will be announced at the 8th HAOMO AI DAY in April this year.
This article is a translation by ChatGPT of a Chinese report from 42HOW. If you have any questions about it, please email email@example.com.