WeRide has launched WePilot AiDrive, its one-stage end-to-end ADAS solution. Developed in partnership with leading Tier 1 supplier Bosch, this milestone comes just six months after both companies brought their two-stage end-to-end solution into mass production.
Unlike the traditional two-stage process of sensing and then decision-making, WePilot AiDrive integrates both into a one-stage architecture, enabling vehicles to ‘see and act’ at the same time – like a skilled human driver. This delivers faster responses, shorter driving routes and greater fault tolerance.
WePilot AiDrive has completed core function validation and is expected to enter mass production and vehicle deployment later in 2025, paving the way for smarter, more efficient and widely accessible assisted driving solutions worldwide.
WePilot AiDrive is engineered to navigate complex road scenarios with ease, including lane changes in heavy traffic; detours around unplanned construction; unprotected turns at intersections; smooth acceleration when following vehicles; and complex interactions with pedestrians, oncoming vehicles and roadside obstacles on narrow roads in urban villages.
It reportedly performs reliably in real-world conditions, smoothly navigating through pedestrians, e-bikes and roadside obstacles at night (pictured below); detecting pedestrians even in strong backlight; handling unprotected turns in complex traffic; executing consecutive smooth lane changes; and maintaining safe distances while executing swift, human-like overtakes in low-light tunnels.
Key advantages
According to WeRide, WePilot AiDrive’s one-stage, end-to-end architecture delivers three key advantages:
Scalable computing power: Runs on high-performance computing platforms and can quickly scale down to mid- and low-power setups via model distillation. Its proprietary middleware decouples algorithms from hardware and foundational software, enabling seamless adaptation across platforms, while its modular Level 2 functions simplify deployment and accelerate integration and mass production.
Adaptability across sensor setups: Supports both pure vision and multi-sensor fusion.
Rapid daily iteration: Uses extensive driving data to automatically generate training labels, continuously improving performance in a cost-effective manner. This strengthens the system’s ability to handle edge cases and long-tail scenarios, delivering on the promise, ‘The more you drive, the better it drives’.
WeRide says mass production and vehicle integration of WePilot AiDrive will begin this year, enabling OEMs and partners to offer “human-like Level 2 autonomy and robust safety capabilities across complex driving scenarios”.
In related news, Tensor, an agentic AI company, has unveiled the world’s first volume-produced, consumer-ready autonomous vehicle, designed from the ground up for private ownership at scale