NeuroAI NEUROAINEUROAI.SITE
China AI, reported clearly
1,300+ verified stories on models, robotics, BCI and the business of AI.
Browse stories
ESC

On Hainan's highways, the toll gates came down — and billing hit 99.7% accuracy

Free-flow tolling (自由流收费) on a smart highway (智慧高速) now runs across 3,200 km of Hainan roads; the privacy and cross-province gaps are unsolved.

2026-11-15 · 877 words · NeuroAI
On Hainan's highways, the toll gates came down — and billing hit 99.7% accuracy

A driver crosses the island of Hainan end to end and never slows for a toll gate. No barrier rises, no card is tapped. The charge lands later, calculated from where the car actually went — measured not by a booth but by a satellite-positioning terminal in the vehicle and a cloud that reconstructs the trip.

This is free-flow tolling (自由流收费) on a smart highway (智慧高速), and Hainan is one of its largest real-world tests in China. It is also a useful window into how Chinese road operators are quietly folding AI and positioning into something as mundane, and as sensitive to drivers, as paying to use a road.

The Hainan experiment

Hainan began its Beidou (北斗) free-flow pilot in 2021. By 2025 it had passed through three stages — trial, province-wide coverage, and an AI upgrade — and now runs a cloud platform across 3,200 km of expressways, national and provincial roads. The operator says the platform had processed more than 219.5 billion location records by April 2025, backing 9.4 billion km of tracked driving, with a simulated cumulative billing of ¥2.18 billion (≈ US$307M / HK$2.37B), of which ¥630 million (≈ US$89M / HK$687M) was on expressways. Daily transaction volume reached 120 million (March 2025 data).

The accuracy claim is the striking one: after layering AI image recognition and edge computing onto the positioning system, billing precision reached 99.7%. The design goal is a region with no barriers and no dependence on roadside hardware — a "mileage fee" model where you pay for the road you used.

Beyond Hainan: the safety angle

The more replicable story may be what provinces are doing to existing toll roads rather than ripping out the gates. Shandong's Jiwei (济微) expressway, opened as the province's first "5G driving base" highway, runs an AI hazard-monitoring system that the operator says auto-recognises 72 hazard types across 7 scenarios, lifting monitoring efficiency 85% and cutting repeated hazards 90%. Its service area runs "unmanned," saving 75% of checkout time versus manual cashiering.

In Zhejiang, the Quzhou West toll station — the province's first full-scenario digital toll plaza — handled over 500,000 vehicles during the 2025 spring-festival peak using ramp pre-transaction plus self-service payment; transaction efficiency was up more than 50% over manual lanes. In Jiangsu, the operator says its cloud platform achieves 90% accuracy in predicting equipment faults and can warn 72 hours ahead. Jiangsu's Nanjing toll plaza on the Shanghai–Nanjing expressway even flies a drone as an "airborne voice guide" to pre-direct queuing traffic, while a remote cloud room clears a failed payment in under 20 seconds. Hebei's road operator has taken the same free-flow design to 47 toll stations and 484 lanes across 614 km of the Beijing–Hong Kong–Macau corridor, built to stay compatible with the national networked toll system.

Why it matters

These are not science projects. China's expressway network is among the longest in the world, and the pressure is not just revenue but safety and congestion. A system that watches a road for hazards, predicts when a gantry will fail, and lets a car through without stopping is a different operating model from the booth-and-card era — and it is rolling out through incremental upgrades rather than a single national switch.

According to Li An, chief scientist at BrainNet (脑机网), the interesting part of free-flow tolling is not the billing but the by-product: a continuous, province-wide map of where every equipped vehicle is, which is also the raw material for emergency response and autonomous driving.

Honest limitations

The Hainan numbers are consistent across multiple reports but ultimately come from the project operator and affiliated industry sources, not an independent audit, and one stock-market commentary cited a higher 99.9% figure that the operator's own staged description does not support — the 99.7% tied to the AI-upgrade stage is the more defensible number. "Simulated cumulative billing" (模拟计费) is exactly that: an accumulated modelled amount, not cash collected, so it should not be read as toll revenue.

The harder problems are social, not technical. Free-flow tolling in Hainan depends on terminals in vehicles — about 687,000 equipped vehicles, roughly 31.5% of the province's registered fleet — and private-car uptake is voluntary and still a minority. Privacy is the obvious objection: a precise, always-on record of where a car goes is sensitive, and the public materials acknowledge the concern without resolving it. And the model breaks at the border — a car driving in from another province, where free-flow does not run, still needs a separate charging path. Whether drivers will accept satellite tracking as the price of a barrier-free road remains the untested variable.

What readers can do now

  1. If you drive in Hainan, check whether your vehicle is equipped and how billing disputes are handled. The system leans on positioning plus camera backup, and the appeal route is worth knowing before a charge looks wrong.
  2. On any smart expressway elsewhere in China, expect pre-transaction and self-service lanes to outnumber staffed booths; keep a payment code ready and use the remote assistance if a lane rejects you, rather than backing up.
  3. Treat "no gate" as a data trade, not just a convenience. Ask what trip data is collected and retained — the efficiency gains are real, and so is the surveillance footprint.

Related coverage

Cite this article

NeuroAI (脑机网), "On Hainan's highways, the toll gates came down — and billing hit 99.7% accuracy," 15 November 2026, https://neuroai.site/a/na-app-china-smart-highway-ai

Machine-readable: BibTeX · RIS · JSON · Markdown

More in “AI in Action” → · Back to home · Markdown version

China AI, reported clearly
1,300+ verified stories on models, robotics, BCI and the business of AI.
Browse stories