XPeng has been anointed the Tesla of China but now it is showing capability that even goes beyond the company fronted by the world’s richest man, Elon Musk. At its Tech Day event in Guangzhou, the EV maker displayed an electric vertical takeoff and landing vehicle (eVTOL) apart from many other goodies like improvements to its ADAS technology and a robotic pet.
Last year, the star of the XPeng Tech Day event was the showcase of its flying car project, and now it has showcased the latest iteration of it which comes with several upgrades. It now gets a distributed multi-rotor configuration, and the design is simpler which will ensure better reliability and safety during the flight. It also gets wheels which support a driving mode like a conventional car as it gets a steering wheel as well.
XPeng revealed that it has completed its maiden flight in addition to multiple single motor failure tests. Beyond this, it has an update to the robotic pet that has a more sophisticated mechanical structure, higher transmission efficiency and a stronger robot actuator. It also has a high-end auto-grade computing platform which is similar to the Tesla Optimus. The robot can also adapt to complex indoor and outdoor terrains like stairs, steep slopes and gravel roads.
XPeng revealed that it was developing a robotaxi network in China based on its G9 electric SUV which became the first commercial vehicle to pass the autonomous driving closed field test led by the Chinese government. It has completed the test in Guangzhou.
All of this was based on the ADAS chops that XPeng has been developing. Its XNGP system has already been announced which includes its ADAS stack with a high-precision mapping solution. Now, XPeng has deployed major hardware updates to the system like a dual LIDAR system, 508 TOPs of computing power and 8-megapixel HD cameras underpinned by XNet which is a software architecture that is a self-evolving AI and data system.
It has also added the XNet neural network which is its first-generation visual perception architecture which is not too dissimilar from what Tesla is doing with the Dojo supercomputer the FSD beta. The idea is to have visual recognition with human decision-making capabilities as the system captures data from the numerous cameras on the cars.
The system can reach 600 PFLOPs in training efficiency — which means a model that usually takes 276 hours will get trained in 11 hours. It is based on the Nvidia Drive Orin X system.
