On 6th Jan, on the G94 route in Guangdong I stepped forward to do an initial test drive of the FSD capabilities of the Xiaopeng P7. Interestingly, this is not the full FSD software, as the approach taken in China is to go piece meal based on scenario and monetizability, of course considering as well possible security and safety concerns.
The video is my simple initial go at testing it:
Observations:
It’s not Tesla. Interestingly the car manufacturer is quick to make this clear that the car is not equipped with software that will enable it do the “wonders” currently boasted by Tesla’s FSD. However, there is a long list of hardware prowess that’s touted by Xiaopeng with their partnership with Nvidia. Recall that Tesla decided to build its own FSD processing unit citing better efficiencies and also given its philosophy that regular optical vision is the better way to deliver autonomous vehicle’s. An assertion I completely agree with by the way.
The car faired pretty well, but decision making was quite primitive. Also I observed some of the key features aren’t taking the full advantage of data that the car is purported to sense. Hopefully this can be available through future updates.
Changing lanes executed well. Ability to sense the speed ahead and effectively maintain the set distance worked too, albeit very robotic by way of abrupt speed increases and decelerations. A couple of other features that were obvious is the inability to reasonably identify all the types of lane markings. Any lane marking that trended into a curve failed the execution of Lane Centering Control as the car got confused. Indicating another programming lock-in or primitive use of the cameras. I used a car with the Xmart 2.1.2 version software, so this isn’t the better version software. There is announcement that during the lunar festival in China the NGP version will be released. NGP being Next Generation Pilot which follows Chinas standard of approaching Autonomous Driving based on scenarios.
Release