Gasgoo Munich - On February 4, 2026, SenseAuto and Dongfeng Motor jointly unveiled the industry's first mass-production solution for generative intelligent driving. Built on SenseAuto's expertise in end-to-end autonomous driving models and driving world models, the system integrates Dongfeng's engineering prowess in mass production, creating a complete loop from R&D to vehicle deployment.
At the heart of this partnership is a one-stage end-to-end architecture, a diffusion generative model, and a training framework combining reinforcement learning with the "KAIWU Driving World Model." By unifying perception, planning, and decision-making into a single model, the one-stage architecture minimizes information loss and latency typical of traditional modular systems, thereby boosting response speeds and decision-making efficiency.
Image Credit: SenseAuto
The diffusion generative model enables the vehicle to generate multiple feasible trajectories in complex scenarios and select the optimal strategy, strengthening its ability to navigate chaotic traffic environments. Combined with the KAIWU Driving World Model and reinforcement learning, the system can continuously train and optimize decisions in a virtual setting, adapting to different driving styles and achieving continuous evolution.
During public road tests in Wuhan, the solution demonstrated its ability to handle highly complex scenarios. It executed smooth, continuous lane changes at tunnel exits where light shifts abruptly and traffic is dense. On narrow, mixed-traffic roads lacking clear lane markings, the system accurately predicted the movements of surrounding vehicles and pedestrians to navigate smoothly. At large, crowded intersections, its decision-making showed a decisiveness and adaptability close to that of human drivers.
Current industry solutions for intelligent driving generally fall into two categories: two-stage end-to-end architectures, where separated perception and planning create bottlenecks in information transfer; and imitation-learning-based approaches, which have limited ability to generalize to new scenarios outside their training data. Through technical innovation, SenseAuto and Dongfeng's generative solution aims to bypass these limitations and build a full-stack intelligent driving capability.
The solution will be deployed across multiple Dongfeng models, providing an evolvable technical foundation for the iteration of its intelligent driving systems. The partnership has further expanded to include cabin-driving integration and the development of high-level intelligent driving platforms, laying the technical groundwork for future Level 3 and higher autonomous driving functions.
