Shanghai (Gasgoo)- On August 18, SAIC-GM signed a strategic partnership with Momenta to conduct profound collaboration in advanced driver-assistance systems (ADAS), according to a post on the latter's WeChat account.
The deal makes Buick's premium new energy vehicle (NEV) sub-brand Electra the first to integrate Momenta's R6 Flywheel Big Model, a reinforcement learning-based algorithm designed to deliver full-scenario assisted driving. The Buick Electra L7 sedan will debut the technology, offering features such as seamless urban Navigate on Autopilot (NOA) and one-touch parking without stopping.
Momenta describes the R6 Flywheel Big Model as the world's most advanced ADAS algorithm and the first reinforcement learning-based system to achieve commercial-scale deployment in China. Unlike earlier imitation-learning models that sought to replicate human driving, the R6 system trains on massive datasets using reinforcement learning, allowing it not only to mimic human performance but to exceed it by continuously refining decision-making through trial and error.
Equipped with this system, the Electra L7 can simulate new driving behaviors, learn from both successes and failures, and improve autonomously. The model’s closed-loop training process ensures continuous iteration, enabling the vehicle to steadily enhance its performance across diverse driving scenarios.
Safety is a central focus. The R6 model has been trained on data distilled from three billion kilometers of real-world driving and refined through 70 million curated data samples of 3-billion-kilometer real-world driving. By exposing the algorithm to challenging "nightmare scenarios," it learns to optimize responses autonomously, developing the ability to generalize across complex environments. This enables the Electra L7 to anticipate and react intelligently to hazards such as sudden lane cuts, hidden pedestrians, or blind-spot obstructions, reducing collision risks.
The system also emphasizes ride smoothness. With the R6, the Electra L7 delivers fluid, human-like handling in city NOA applications, ensuring no interruptions in navigation. The vehicle can manage tight streets, lane borrowing, and narrow passes with greater precision, adjusting acceleration and braking seamlessly. Even toll station passages—long considered a bottleneck for NOA systems—are handled efficiently, with the Electra able to detect barriers and drive "blind" without lane markings, moving through entry, gate, and exit in a single, uninterrupted motion.
Efficiency is further enhanced through one of the industry's first "non-stop one-touch parking" features. The Electra L7 can identify open spaces while moving, plan the optimal trajectory, and execute parking smoothly without pausing for manual setup. The vehicle model also supports advanced functions such as customizable modes, robust coverage of complex scenarios, and memory-based parking, enabling it to tackle even challenging mechanical garages with precision. Combined with improved lane-change timing in traffic and more accurate navigation at complicated intersections, the vehicle significantly boosts travel efficiency.
