In-depth report on the automotive industry: comparative analysis of autonomous driving technology paths (with report)

report summary

At present, the autonomous driving system has been installed one after another, and the leading enterprises can basically achieve the overall L2.5 level and the L3+ level of some functions. The level of automobile intelligence and automation has been significantly improved.

In the process of the development of the autonomous driving system, technical bottlenecks have been encountered at both the perception and decision-making levels. In terms of perception, it is mainly visual perception and radar sensing, and these two perception directions have their own shortcomings, including but not limited to insufficient resolution, large influence by external environment such as climate, high cost, etc. It is difficult to deal with more complex road conditions, or it is difficult to guarantee safety; in terms of decision-making, the hardware side has the problem of insufficient computing power or high power consumption, and the software side has inefficiency in the case of an increasing amount of data. However, the safety of the vehicle cannot be guaranteed after the further expansion of the data volume in the future.

At present, there are a large number of manufacturers engaged in the development of autonomous driving systems and they are widely distributed, and since the entire system requires multiple links from hardware architecture to software writing to car company verification, the entire industry chain involves a lot of content, but in general it can be The realization of mass production and loading on the road is basically still at the L2.5 level, and some functions can reach L3 or above. There is still a long way to go before the highly automated driving of L4 and L5.

From the perspective of the technical path, there is not much difference on the software side. In essence, they all rely on machine learning algorithms to achieve iteration by combining actual drive test data and simulated drive test data. The main difference on the hardware side is whether to use lidar: due to its excellent performance, in order to ensure the safety of autonomous driving, most manufacturers choose to carry lidar, including Waymo, Volvo, GM, etc.; and due to high costs, Based on the consideration of commercial mass production, Tesla does not use lidar, but uses a visual method as the main method, supplemented by ultrasonic and millimeter wave radar to build its perception module. We believe that from the perspective of prioritizing safety, lidar will still be one of the most important sensors for autonomous driving systems in the future, and its current high cost will be significantly reduced under the multiple effects of technological progress and scale effects, making its sufficient economical.

Considering the final mode of autonomous driving, the integration of vehicles and roads is the final ideal state, but this will also be an extremely long development process. During this process, we believe that the digital economy development strategy promoted by the state will continue to be the driving force for vehicles and roads. Support for collaborative development.

To view the full report click here

Note: The articles on this site have not been posted and shared by netizens or institutions unless they are marked as original. If there is any need for publicity or infringement, please contact [email protected]

This article is reprinted from:
This site is for inclusion only, and the copyright belongs to the original author.

Leave a Comment

Your email address will not be published.