How Didi’s AI‑Powered Map Engine Redefined Ride‑Sharing Navigation

Didi, after six years of rapid growth, has become a global leader in shared mobility, relying heavily on its map platform to support the entire ride‑hailing workflow—from order dispatch to settlement. Accurate, timely map data and AI‑enhanced services such as route planning and ETA estimation are essential for smooth trips.

During the ArchSummit global architects summit on July 7 in Shenzhen, senior expert engineer Zhu Zhiqing presented Didi’s practical explorations and AI applications in its map system architecture. Leveraging massive real‑time travel data, Didi’s map now offers core services like path planning and ETA , while integrating AI to remodel classic map problems for higher accuracy and relevance.

In path‑planning, traditional models have been transformed into AI‑driven solutions. By learning from billions of positioning trajectories daily, the system recommends more reasonable routes, reducing deviation rates to industry‑leading levels and improving overall route efficiency by about 3%.

For ETA modeling, the system’s perception capability has been enhanced to provide real‑time updates every two minutes, and machine‑learning models further increase prediction accuracy.

To cope with increasing business and technical complexity, Didi adopted a "model + feature" organization. An engine control layer now manages strategy module calls and experiment controls, accelerating business iteration speed.

On the model‑iteration side, a strategy plugin system decouples strategies from the core architecture, lowering deployment labor costs and breaking the tight coupling that previously hindered rapid experimentation.

Regarding data freshness, Didi built a global‑plus‑local data update framework with version management, shortening the back‑track cycle of massive offline data and meeting high‑availability, high‑performance requirements.

Overall, Zhu emphasized that Didi’s strong internal drive—massive high‑quality travel data, efficient feedback loops, and a thriving AI ecosystem—has been crucial for breakthroughs. The company continues to pursue faster data‑update pipelines, more agile strategy iteration, and mature model capabilities, aiming to further transform intelligent mobility.