How Baidu’s Large‑Model ‘Yuanji’ Is Transforming Traffic Policing in China

Background

In the era of large‑model AI, intelligent question‑answer (Q&A) products are expanding into new sectors, and traffic management has become a recent focus.

Digital Police “Yuanji”

Shijiazhuang traffic police officially launched the always‑online digital officer “Yuanji” (also known as “Shi Xiao An”). The system is built on Baidu Cloud’s Qianfan large‑model platform together with a traffic‑industry‑specific model, giving it a “super brain” capable of handling complex sentences, multi‑turn dialogue, and voice interaction.

Capabilities and Benefits

Yuanji provides human‑like conversation, memory across turns, and can answer a wide range of traffic‑related queries. It addresses common pain points such as “no answer,” “incorrect answer,” and “unreadable response” by leveraging language understanding, intelligent reasoning, and data analysis.

Operational Results

After a test launch in May, the assistant handled more than 150 frequent questions about illegal parking, driver licensing, and other topics. It achieved a 98 % response efficiency and a 95 % factual accuracy, earning strong approval from both police staff and the public. Users can access Yuanji via the official WeChat account, a WeChat mini‑program, or by scanning a QR code.

Industry Impact

Compared with earlier chatbots, the large‑model‑enhanced system improves three key areas: anthropomorphic interaction, higher accuracy and professionalism through a locally built knowledge base, and integrated cross‑system knowledge that enables multi‑dimensional analysis of complex issues.

Broader Deployments

The Baidu Cloud traffic‑industry Q&A platform has been rolled out in Baoding, Shijiazhuang, and Xi’an. In Baoding, it answered over 100 categories of questions with 99 % response efficiency and 95 % factual accuracy. In Xi’an, it covered the top 200 public queries with 95 % factual accuracy and 97 % response efficiency, and it links feedback directly to internal departments for rapid issue resolution.

Technical Foundations

Success stems from the combination of Baidu’s general large model—renowned for Chinese language generation—and a traffic‑industry model trained on 12 national traffic laws, 32 policy documents, more than 60 standards, and roughly 9,000 local regulations specific to Shijiazhuang. Role‑configuration features allow the system to adopt a personalized tone, greetings, and sensitivity settings, creating a more natural user experience.

Strategic Cooperation and Future Outlook

Since September 2023, Baidu and the Shijiazhuang traffic bureau have cooperated on signal‑control optimization, situational analysis, and road‑hazard detection. Baidu’s ACE intelligent‑traffic engine 3.0 already serves over 200 cities, and version 4.0 is slated for release at the 2024 Cloud Conference, promising further AI‑driven innovations for smart transportation.