I’m writing this from Hangzhou, in the middle of the Apsara Conference 2019 (云栖大会). A few days ago I was awarded MVP of the Year 2019 at the MVP Global Summit, and since then I’ve been walking the exhibition halls trying to process everything China is doing with AI on real streets, in real hospitals, in places where normal people live.
This year’s topic is AI, and the ET City Brain is the main focus. Hangzhou is Alibaba’s home city, and it is also the city where City Brain first went live back in October 2016. Now we have 3 years of data to see how it is working.
If you read my post about the Hanguang 800, you know Alibaba already has its own AI hardware. That chip exists because China has decided AI is not optional. The government’s “New Generation Artificial Intelligence Development Plan” sets the goal to become the world leader in AI by 2030. Ambition? Maybe, but we are talking about production deployments.
City Brain is the perfect example. It belongs to the broader ET Brain family (Alibaba Cloud’s umbrella brand for applying cloud computing, big data, and AI to verticals like traffic, manufacturing, agriculture, and healthcare). The idea is to ingest city data (cameras, traffic lights, GPS from taxis and mapping apps, telecom signals, weather feeds, you name it), run it through deep learning models on Alibaba’s Apsara computing platform, and feed the results back into the city in real time.
The video above is an introduction to ET City Brain by the Alibaba Cloud team. It walks through the architecture and shows real dashboards built for Hangzhou (congestion maps, intersection health scores, ambulance priority scheduling, bus route optimisation, etc). Worth watching if you want a technical tour.
Hangzhou is a major city with millions of commuters, a historic centre, highways feeding into the Yangtze Delta, and the kind of traffic that makes European capitals look manageable. When City Brain launched here, sceptics asked the obvious question: can AI really do better than traffic police with radios? Well, it looks like yeah, it can!
In the Xiaoshan district, City Brain took over traffic signal timing. Not fixed cycles based on a spreadsheet from the 1990s, but dynamic plans from live data. The reported results:
- Daily commutes shortened by roughly three minutes
- Average travel speeds up 15%
- Peak-hour congestion down 9.2%
Hangzhou’s downtown area feeds 3,500 traffic cameras into the platform and City Brain analyses it in real time (vehicle detection, incident recognition, queue length at intersections) with accuracy close to 95% as reported from Alibaba DAMO Academy.
City Brain also optimises public transport. In Suzhou, a pilot adjusted bus schedules and routes based on real demand data and moved 17% more passengers without adding a single bus. Same fleet, smarter routing.
Treating emergencies as emergencies
In a normal congested city, when you hear a siren, cars ahead do not always move or sometimes move but slowly. With City Brain, when an ambulance requests a route, the system calculates the shortest path, identifies every intersection along the way, analyses queue lengths from video feeds, and times the green/red lights so the ambulance arrives to cleared roads.
In Xiaoshan, emergency vehicle travel times dropped by 50%, with rescue teams arriving roughly seven minutes earlier. This is what impressed me the most. For a car to sit in traffic is just annoying, but for an ambulance to sit in traffic can be a life-or-death situation.
More from my Apsara Conference 2019 coverage: conference overview, MVP of the Year, Hanguang 800 and Chinese AI.