Pony Zhixing joined hands with Nvidia to launch a new automatic driving system, and Toyota will take the "first train"

The partnership between self-driving company Xiaoma Zhixing and chip manufacturer NVIDIA has further deepened.

On January 20, Xiaoma Zhixing announced the launch of a self-developed vehicle specification-level computing unit solution, equipped with Nvidia DRIVE Orin (SoC) system-level chips, which is promoting the mass production of the next generation of autonomous driving software and hardware systems to accelerate the large-scale deployment of L4 level autonomous driving technology, and the new computing unit will begin mass production by the end of 2022. At the same time, Xiaoma Zhixing released the sixth generation of automatic driving system solution, the first model equipped with the system is Toyota "Xena SIENNA" Autono-MaaS vehicle, will start road testing in China this year, in the first half of 2023 into the daily operation of autonomous driving services (Robotaxi).

As early as 2017, Xiaoma Zhixing established a partnership with NVIDIA and used the NVIDIA DRIVE platform for the first time. Since May 2021, Xiaoma Zhixing has begun to cooperate with NVIDIA to develop Orin-based autonomous driving computing units, which has accelerated its hardware development process. Based on the cooperation with NVIDIA, Xiaoma Zhixing has launched a variety of computing unit solutions with different chip configurations, which can meet the technical requirements of the development of autonomous passenger cars and autonomous trucks at the same time.

"The sixth-generation autonomous driving system aims at vehicle-level mass production from the aspects of modeling design, parts development and selection, software and hardware coupling, safety redundancy and system assembly and production." Peng Jun, CEO of Xiaoma Zhixing, said that the new generation of automatic driving software and hardware systems continue to follow the technical route of deep integration of multiple sensors, and for the first time, solid-state lidar has been adopted on a large scale, with a total number of 23 sensors, of which the resolution of self-developed signal light recognition cameras has been increased by 1.5 times. In addition, the new generation of systems has multiple layers of redundancy, including complete software, computing platform and vehicle platform redundancy system, which balances the coverage of the scene and the safety processing mechanism to the greatest extent. In an emergency, pull-over or safe parking can be achieved to minimize safety risks.

A research report released by Anxin International Research Institute shows that lidar can be divided into three categories according to the technical path subdivision: mechanical, hybrid solid-state, and solid-state; among them, mechanical development is earlier and technological maturity is high; solid-state is better than mechanical in performance and cost, but there is still a breakthrough in technology. In the future, solid-state lidar will replace the existing mechanical lidar, because solid-state lidar can solve the problem of high material cost + high mass production cost faced by mechanical lidar.

At present, the realization of L4 level automatic driving function is the direction of the industry's efforts, and its requirements for computing power are higher. The industry believes that the current hardware is not the bottleneck of the promotion and popularization of L4-level autonomous driving, but whether L4-level autonomous driving products can be promoted and applied depends on whether the cost and laws and regulations support it. Cai Zheng, general manager of the automotive business department of Texas Instruments (TI) China, believes that the core chip that meets the L4 level not only has a higher cost of its own, but also its auxiliary power supply system, heat dissipation system and the entire system cost are also increasing, and the continuous improvement of these costs may be the most important factor limiting the market expansion rate of L4 level autonomous driving products in the future.

In Peng Jun's view, scale and unmanned will be a key step in the development of autonomous driving technology, and the industry is moving in these two directions. The rapid maturity of the industry supply chain can further reduce the cost, and the rapid reduction of the cost of sensors and the continuous improvement of stability will lay the foundation for the research and development of the next generation of autonomous driving software and hardware systems. Compared with the previous generation of computing units, the new generation is expected to increase the hash rate by at least 30%, reduce the weight by at least 30%, and reduce the cost by at least 30%.

"We are on the eve of the widespread adoption of autonomous driving products in the consumer market, gradually leaping from testing and verification to product applications. More and more mass-produced models are beginning to use assisted driving technology and sensor configuration as selling points for dissemination; OEMs are beginning to enter the Robotaxi track; logistics companies are beginning to see the value of autonomous driving technology, and cooperation with autonomous driving companies is beginning to be reached. At the same time, in practical applications, the boundaries of autonomous driving technology at the L2 to L4 levels begin to blur to a certain extent in terms of hardware redundancy and functional definition. Peng Jun said.

At present, the self-driving test mileage of Xiaoma Zhixing has exceeded 10 million kilometers. For the next five years, Peng Jun said that he will continue to focus on several major businesses with the "virtual driver" autonomous driving platform as the core, including robotaxi, smart logistics business (Robotruck) and other technology areas that reshape transportation.