Hong Kong authorities are expanding autonomous-vehicle tests and opening more roads for trials as Baidu scales its robotaxi operations in the city.
This expansion represents a push to modernize urban transit and advance the technical capabilities of driverless systems within complex city environments. By increasing the scope of trials, the city aims to transition from limited pilot programs to fully remote-controlled commercial services.
Baidu recently demonstrated its sixth-generation autonomous vehicle on a route on Lantau Island. The vehicle completed a seven-minute driverless run [3] before a manual takeover was required once the car left its designated geo-fenced zone.
The company is now transitioning its robotaxis toward fully remote-controlled operations. This shift allows human operators to manage vehicles from a distance, reducing the need for on-board safety drivers while maintaining oversight.
To support these technological leaps, the government has increased the number of pilot licenses available for testing. In June 2024, authorities issued seven pilot licenses [1], which currently cover a total of 63 vehicles [2].
These licenses allow companies to test various sensors, and software iterations on public roads. The increased volume of vehicles enables more diverse data collection, which is essential for refining how AI handles the unique traffic patterns of Hong Kong.
The current trials focus on specific zones, but the goal remains the expansion of these boundaries. The ability to move beyond geo-fenced areas without immediate manual intervention remains a primary hurdle for the industry.
“Hong Kong authorities are expanding autonomous-vehicle tests and opening more roads for trials.”
The transition toward remote-controlled operations suggests a middle-ground strategy between human-driven and fully autonomous transit. By utilizing a fleet of 63 vehicles across seven licenses, Hong Kong is creating a high-density testing environment to solve the 'edge cases' of urban navigation. The requirement for a manual takeover after leaving a geo-fenced zone highlights that while hardware is advancing, the software's spatial autonomy is still limited to predefined maps.
