Haomo.AI Releases An Autonomous Driving Generative Large-scale Model
On April 11, Haomo.AI, a Chinese AI technology company, launched “DriveGPT,” an autonomous driving generative large-scale model. Zhang Kai, the Chairman of the startup which is supported by automaker Great Wall Motor, stated that a true AI era has arrived and generative large-scale models will be key to the development of autonomous driving.
Gu Weihao, the CEO of Haomo.AI, announced that DriveGPT can enhance the cognitive decision-making model of autonomous driving through reinforcement learning from human feedback (RLHF) by incorporating actual human driving data. DriveGPT currently possesses three primary functions: generating multiple scene sequences simultaneously, predicting future vehicle trajectories, and producing decision reasoning chains. With a unified generative framework, it can perform various tasks ranging from planning and decision-making to inference.
Gu explained the implementation process for DriveGPT in detail. The team began by training an initial model using mass-produced driving data during the pre-training stage. They then trained the reward model and used reinforcement learning to continuously optimize and iterate on the initial model, resulting in a cognitive decision-making model for autonomous driving that underwent continuous optimization.
Mocca DHT-PHEV under Great Wall Motor will be the first model to have DriveGPT’s intelligent driving capability. DriveGPT has already opened up to the industry and is currently collaborating with a limited number of customers, which include Beijing Jiaotong University’s School of Computer and Information Technology, Qualcomm, Bytedance’s Volcano Engine, Huawei Cloud, JD Tech, NavInfo, and Intel.
“2023 is expected to be a year of rapid growth for the intelligent driving industry,” stated Zhang. “Intelligent driving products will experience an explosive surge, starting this year. Major players in the industry, such as Huawei and XPeng, are launching their products one after another. This means that urban navigation-assisted driving will become more accessible to users.”
Moreover, the range of smart driving products will expand to include mid-to-low-end car models, while automatic logistics in supermarkets and express delivery services will be enhanced.
SEE ALSO: Autonomous Driving Startup Haomo.AI to Deliver Urban Driving Assistance Product in September
Zhang forecasts that by 2025, over 70% of passenger cars will have advanced driver assistance systems. The autonomous driving industry is currently shifting from a software-driven era with limited data and models to a data-driven era characterized by extensive data and parameter models. As such, measuring user usage frequency and satisfaction will be crucial in determining the competitiveness of intelligent driving products.