TMTPOST -- A report on Thursday sheds light on Tesla Inc. is pushing toward its real self-driving software in China.

Credit：Tesla

Tesla is partnering with Baidu to improve its advanced driving assistance (ADAS) system in China after recent software update drew criticism among Chinese customers, Reuters cited anonymous sources.

In recent week, a group of engineers in Baidu’s mapping team has been deployed to Tesla’s Beijing office to enhance the integration of Baidu’s navigation map information, such as lane marking and traffic light signals, with Tesla's Full Self-Driving (FSD) Version 13 software, according to the sources. It was reported that the initiative aims to make FSD Version 13 better understand Chinese roads with more accurate and updated mapping information.

The sources didn’t reveal how many Baidu engineers are working on the improvement of Tesla’s system, neither did a financial value to the collaboration. Tesla and Baidu didn't respond to requests for comment on the reported matter yet.

The report showcased Tesla is striving to launch services powered by autonomous technology compliant with Chinese regulatory requirements amid intense competition in China, its second biggest market as well as the world’s No.1 electric vehicle (EV) market.

Tesla still can not release FSD system comparable to that in U.S. due to data restrictions and local regulatory constraints have hampered system training, while Chinese rivals like BYD Co., Ltd. and Xpeng Inc. have offered cheaper vehicles and similar self-driving similar to Tesla without additional fee.

Tesla on February 25 updated its autopilot software in China. The new version have a new feature called Urban Road Autopilot Assistance (URAA). On controlled-access highways and urban roads, URAA can guide vehicles according to navigation routes, assisting with entering and exiting highways, navigating intersections, and recognizing traffic lights to perform actions like going straight, turning left or right, or making U-turns, according to a notification on Tesla’s application. It added that the in-car camera will monitor driver attention and send alarm to help driver focus on the road when driving.

FSD is a suite of driving-assistance technologies developed with generative artificial intelligence (AI) to handle more complicated traffic conditions whereas Autopilot handles more routine conditions. Reuters reported that the upgrade in China left owners disappointed for they found the URAA feature fell short of Musk’s promises of self-driving.

The latest over-the-air (OTA) update of driving assistant software costs Tesla car owners RMB64,000, and Tesla China officially change the name of FSD feature from “full self-driving” to “intelligent driving assistance”, the customer service told Chinese news media outlet YiCai. The customer service staff made it clear that the update is different from the FSD that is available in the United States, and the car with the updated software can not drive itself without human intervention in some cases.

The new feature is less advanced than those hit in U.S. market due to insufficient data training on Chinese roads and traffic rules, Reuters cited a person with direct knowledge of the matter.

Tesla AI Team said in last September it will release FSD in China and Europe in the first quarter of 2025, pending for regulatory approval.

On an earnings call earlier last month, Musk said Tesla is going to launch “unsupervised full self-driving as a paid service” in Austin, Texa, in June. That means Tesla plans to start its paid robotaxi service powered by FSD (Supervised) software in Texas in mid-2025. Musk didn’t update the FSD rollout timetable in China, though listed challenges Tesla is facing including data collection for AI to train the FSD software.

Musk described Tesla’s situation as a quandary for the Chinese government doesn't allow Tesla to transfer training video outside of China, and the US government won’t let it do training in China. Tesla’s way to overcome it is to create a simulation. Musk told analysts the company searched videos of streets in China that are available on the internet to understand, then fed that into its video training and put it into a very accurate simulator. It also had to made the simulator replicate complex traffic rules like what time of the day bus lanes are available for automotive transportation as well as very complicated road conditions of the lanes.