Page 13 - EE Times Europe Magazine - June 2025

EE|Times EUROPE 13

WeRide’s Robobus Reaches Barcelona as Level 4 Ambitions Accelerate

WeRide has just launched paid commercial Level 4 robobus service in central Guangzhou. (Source: WeRide/Guangzhou Daily)

Each vehicle type uses similar algorithms, so different sensor config- As a case in point, the first robobus version had its top speed
urations can navigate urban environments autonomously. The vehicles software-locked to just 15 km/h on private roads, but the latest version,
rely on a host of sensors; GNSS, inertial measurement units (IMUs), with upgraded LiDAR hardware and software algorithms, has clearance
LiDAR, radar, cameras, and a custom sensor board. “[Our] modular to reach 40 km/h on public streets. “Our prediction models, trained
sensor suite can be configured for different vehicles, sharing more than with extensive real-world data, facilitate smooth interactions … accu-
90% [of the] parts,” Han said. rately predicting behaviors in complex scenarios,” Han said.
The robobus itself contains up to five LiDAR sensors for 360° envi- But having amassed vast swaths of real-life data, does the WeRide
ronment mapping and obstacle detection. Custom RGB and fisheye lens platform pose data privacy concerns? Han offered this assurance:
cameras provide data on lane detection, traffic sign and signal recog- “After personal information of traffic participants, such as license
nition, and pedestrian and vehicle classification. The fisheye cameras plate numbers or human faces, are picked up by the sensors, they are
observe obstacles that are close to the vehicle, while the other cameras automatically desensitized when uploaded to our cloud-based data
serve as long-range sensors for, say, detecting traffic lights in the dis- platform. Any original video clips that contain relevant personal infor-
tance. Radar on the vehicle bumper also detects objects and can support mation are then removed.”
the LiDAR and other cameras in adverse weather and low-light condi-
tions. Vehicle localization comes from both the GNSS and IMU sensors. HYPERVISION BOOST
“Our positioning technology combines multisensor fusion and 3D Even with WeRide’s AI-led sensor architecture keeping a close eye
high-definition maps to provide precise, real-time localization,” Han on the road, the company partnered with France-based automated
said. “[We have] reliable positioning in diverse environments, including mobility operator beti to add a hypervision remote supervision
tunnels, bridges, and urban areas surrounded by skyscrapers.” platform for extra support. With hypervision, trained operators, or
hypervisors, monitor the autonomous vehicles in real time. “Our
AI AND AUTONOMY hypervisors can’t drive our shuttles remotely, but they can react to
In any autonomous vehicle, AI connects all sensors so it can perceive system alerts and stop a vehicle,” Han said.
its surroundings, make decisions, and navigate complex driving situa- So where is WeRide headed next? Worldwide expansion is a priority,
tions while on the move. WeRide has developed deep-learning models Han said. Market research firm CIC predicts the size of the global
for object perception; they include sensor fusion algorithms, as well as autonomous driving market will reach $1,724 billion by 2030—which
prediction, planning, and vehicle control. “For example, our perception Han said would amount to a 98% rise on a CAGR basis from 2021. He
model adapts to various sensor setups and vehicle types, while plan- added that L4 autonomous driving in particular is expected to grad-
ning algorithms are designed for general urban scenarios,” Han said. ually dominate the global market worldwide, accounting for 88% of
To avoid accidents, the perception model recognizes and tracks overall revenues.
objects in real time, fusing LiDAR and camera vision maps to reconcile “We intend to build on our technological and business milestones to
all the gathered data and generate 360°sensor coverage. Meanwhile, advance toward full commercialization across all [vehicles],” Han said.
the prediction model learns how other vehicles behave on roads to gen- Toward that end, in May, WeRide released its first fare-charging
erate probable trajectories. Planning algorithms use neural networks robobus services in its hometown of Guangzhou, then days later
and game theory to anticipate humanlike driving behaviors, also gen- revealed plans with mobility platform Uber to launch robotaxi services
erating and optimizing trajectories. And a control module translates across 15 cities globally in the next five years. The partners already
driving trajectories into action, ensuring precise vehicle maneuvers. operate together in Abu Dhabi and are about to do the same in Dubai.
Thus far, WeRide has amassed more than 15 million kilometers of “By [using] the platform as our foundational model and business
autonomous driving mileage. According to Han, gathering real-world backbone, we are now poised to seize industry tailwinds … and redefine
data along the way has been key to vehicle development, as the deep- industry horizons,” Han said. ■
learning models are continually trained with new data via the cloud-
based platform. Rebecca Pool is a contributing writer for EE Times Europe.

www.eetimes.eu | JUNE 2025