Chinese Delivery Robots Go Night Shift: Where Autonomy Breaks Down

Chinese developers of autonomous couriers are seriously targeting the night shift. According to Bloomberg Tech, the industry believes that after-dark delivery opens up a large market, but only if the technical difficulties can be overcome. The original source gives no public figures on revenue, fleet size or the share of night trips. So below we look at why companies are drawn to the night and where the vehicles run into trouble.

Why night makes sense

During the day, driverless vans and small delivery robots share the road with scooters, pedestrians, taxis and cars double-parked. Each of these episodes forces the system to be cautious and slows the trip down. At night the roads are clearer, speeds are higher and routes are simpler.

There is also the economics. A vehicle with an expensive sensor suite and onboard compute pays for itself only at high utilization. If it sits idle from evening to morning, depreciation keeps growing while no revenue comes in. Night trips can fill that idle time: shuttling goods between warehouses, restocking stores, late-night orders of food and other goods. For human couriers this is an inconvenient shift; for a robot it is an ordinary one.

What breaks in the dark

The engineering problems start with perception. Cameras in low light produce noise and lose detail. Oncoming headlights wash out the frame, and wet asphalt adds glare. Lidar does not depend on lighting, but it has its own limits: rain, fog and reflective surfaces.

There is also a data problem. If a recognition model was trained mostly on daytime recordings, at night it faces a different data distribution. A person in dark clothing, a cyclist without a light, an unlit barrier gate, a box in the roadway: these are all rare scenarios, and rare scenarios are where autonomous systems make the most mistakes.

The main bottlenecks:

• camera image quality in low light and heavy glare;
• detection of people and small obstacles without supplemental lighting;
• a shortage of labeled night data for training and validating models;
• unpredictable human behavior: drunk pedestrians, teenagers on scooters, animals;
• vandalism and cargo theft when there are no passersby or security nearby;
• engine and speaker noise in residential neighborhoods.

Regulation and money

Technical readiness is not enough. Chinese cities grant permits for driverless vehicles by zone and vehicle type and, judging by the logic of the market, often with time-of-day restrictions. Expanding to night will require either new permits or safety evidence that convinces officials. That means more test kilometers and intervention statistics than daytime operation needs.

At the same time, night operation is not necessarily more expensive to run in itself. If the system works reliably in the dark, the cost per delivery falls, because the fleet is used more and the roads are clearer. But if a remote operator has to step in often, the benefit disappears.

What to watch next

For logistics companies, the key question is not whether a robot can drive at night, but how many night kilometers it covers without human intervention. It is worth asking suppliers for separate day and night statistics rather than blended averages.

Investors and analysts would do well to track three metrics: the share of night trips in total volume, the number of manual interventions per thousand kilometers after dark, and the list of cities where night permits have actually been issued. Developers need to collect their own night datasets and add rain, glare and rare objects to simulation. Whoever first proves stable performance in the dark will gain a noticeable advantage in fleet utilization, and therefore in delivery pricing.