In recent years, many developers, including me, have devoted themselves to the field of smart cars. In fact, whether they are beginners or masters, they need to learn a lot of knowledge and skills after contacting the smart car development industry.
I have come into contact with Baidu's Apollo open platform recently, and it feels very easy to use. It just so happens that Baidu has also launched a new version of the Apollo open platform 8.0. Some views and perceptions of Apollo Open Platform 8.0.
Lower development threshold, short deployment time, flexible and easy to expand
The new Apollo Open Platform 8.0 this time can be said to have undergone an all-round upgrade, and has made great progress in terms of framework and capabilities. From the framework point of view, Apollo Open Platform 8.0 has been upgraded from the previous technology-oriented layered architecture to a new architecture that combines technology and ecological layering. The bottom layer is the hardware device layer, which is composed of the previous hardware development layer and vehicle certification layer. Attract ecological partners through open standard protocols and certification to provide developers with a wealth of options for autonomous driving equipment.
Its upper level is the software core layer, which evolved from the previous open source software layer and provides the minimum software subset for building an Apollo-based autopilot system. Provides PnC simulation and debugging capabilities based on local HMI Dreamview, which greatly improves the efficiency of perception and PnC research and development. Further up is the software application layer, which provides developers with new self-driving module expansion capabilities and expansion methods. The top level is the cloud service layer, which has added service capabilities such as practical training and model training, and has upgraded simulation capabilities.
The new architecture has laid a good foundation for the advanced technology of Apollo Open Platform 8.0. Through the architecture setting of "hardware-software core-software application-cloud service", Apollo Open Platform 8.0 has formed a closed-loop integration of software and hardware. At the application level, the hardware level can be efficiently coordinated with the software level, and at the development level, it provides developers with an excellent development and testing environment, allowing the entire platform to have stronger capabilities.
As far as I am concerned, this structure is a bit like Apple's philosophy. Both MAC OS and IOS systems are closely related to hardware. Logically speaking, the architecture of Apollo Open Platform 8.0 ensures the safety of hardware operation. At the same time, the system can be embedded into the hardware according to the different levels of the application layer and the core layer, so that the operation of the software is more stable and smooth, and at the same time, it can also maximize the performance of the hardware without excessive stacking. Many high-performance settings and performances can be achieved.
As for what I just said about enabling the platform to have stronger capabilities, I want to understand clearly that I think we should first analyze the advantages of Apollo Open Platform 8.0 in terms of ease of use and scalability. Apollo Open Platform 8.0 has greatly improved the ease of use and development efficiency of Apollo engineering by covering the entire process of software package management mechanism, perception and PnC development. It can lower the development threshold for individual developers or small studios and provide more development possibilities. . In the future, with the increase in the frequency of automotive OTA upgrades and the iterative speed of vehicle products, the ease of use of Apollo Open Platform 8.0 will show a more efficient side in product development.
In addition, the scalability advantage of Apollo Open Platform 8.0 is also a highlight of this upgrade. Apollo Open Platform 8.0 has introduced a software package management mechanism to solve the problem of tight coupling between compilation and release of the previous Apollo project, greatly reducing installation and engineering costs. The threshold of expansion greatly shortens the installation and deployment time, and is more flexible and easy to expand. It can shorten the development cycle and reduce the utilization rate of human resources. It supports secondary expansion with a high degree of freedom. It can not only share and brainstorm collaborative work, but also build its own scene project for in-depth customized development.
In this regard, I think Baidu's ideas are very reliable. For example, MAC OS and IOS give developers a complete and highly integrated toolkit including Xcode and Swift. Developers don't need to work hard to build it. Its own private environment guarantees development efficiency, stability, and collaborative operations, and also ensures the compatibility of development projects with hardware. Apollo Open Platform 8.0 is a bit like a combination of these toolkits. Of course, this analogy is not It is very appropriate, but it can be said that it is roughly similar. All in all, it makes it easy to get started, and anyone with a little foundation can master it proficiently.
In addition, the open source and community-based advantages of the Apollo Open Platform 8.0 are also in line with the habits of us developers. On 8.0, you can solve problems at one stop. Everyone’s wisdom is gathered together. In any case, it is better than fighting alone. Even individuals have the ability to contract some development tasks to earn extra money, and learn more industry and development knowledge. , Accumulate richer development experience.
The accuracy of the framework is improved, and the efficiency of debugging and verification reaches a new high
The advantages of the new architecture of Apollo Open Platform 8.0 are actually reflected in the improvement of development quality. The intelligent assisted driving function of smart cars and the future automatic driving function still have many technical difficulties, such as low precision, limited scenarios that can be simulated during the development process, not enough to cover the real scene, and long development cycle requires a lot of cost, etc., and Apollo The emergence of Open Platform 8.0 has come up with solutions to these pain points.
For example, Apollo Open Platform 8.0 introduces three new models based on deep learning, and has a new upgrade in the perception framework. In terms of Lidar perception, the CenterPoint laser point cloud obstacle recognition model is implemented. Compared with the MaskPillars model released in 7.0 , CenterPoint does not need to manually set the Anchor size, but returns the size, direction and speed of the object based on key point detection, providing higher accuracy in complex scenes with various object sizes.
In terms of camera perception, the CaDDN visual obstacle recognition model is implemented. By predicting the depth distribution of each pixel in the image, combined with the bird's-eye view projection, the accuracy is higher than that of the SMOKE model released in 7.0. At the same time, it has added a visual BEV perception model PETR, which innovatively combines 3D coordinate information with image features to realize 360° obstacle perception based on vision, and achieves a good balance between speed and accuracy.
In addition, the upgrade of the Apollo Open Platform 8.0 perception model has effectively improved the accuracy on the basis of the original, and also introduced 360° visual BEV perception to enhance the visual perception ability. It can help developers achieve high-precision function development, effectively improving efficiency and reducing costs. In addition to using the models that come with the platform, developers can also expand more models based on their own needs. The development environment is more free and The verification process is assisted by visual tools, which further reduces the difficulty of development.
It is worth mentioning that Apollo Open Platform 8.0 also adopts a new PnC tool chain, and integrates PnC simulation debugging tools in the local Dreamview, which greatly improves development efficiency. The system supports local simulation and debugging, and can simulate vehicle driving and reproduce various scenes through the simulator of dreamview. Moreover, it also supports the cloud simulation scene management function. Developers only need to create scenes and obstacles that meet their own test simulation needs to perform system simulation verification. One-time modeling can be reused. The scene also supports classification management and can be verified to More scenarios, and the verification process is simpler and more efficient.
In order to improve efficiency, the platform introduces the Studio plug-in in Dreamview, through which the scenes and dynamic models on the Studio can be conveniently downloaded locally. In this way, the PnC debugging efficiency is more than doubled. In this ecology, developers will be able to devote themselves to technology development, and all other problems can be solved by Apollo Open Platform 8.0. It can be said that Apollo Open Platform 8.0 is a fully upgraded, highly integrated and intelligent development platform, which has an extremely leading position in the industry and is a good partner for developers.
Summarize:
In fact, to sum up, the overall feeling of Apollo Open Platform 8.0 is still "friendly". It is friendly and allows more cross-industry developers to participate, and its overall accuracy and development and debugging efficiency have been greatly improved, which solves many pain points and difficult problems in the industry, breaks through the limitations of technical bottlenecks, and at the technical level It is also advanced enough to be able to develop more competitive products. This is also very important for developers, and it can be said to be a development tool.
Finally, I would like to share with you that after the launch of Apollo Open Platform 8.0, a brand new community - Apollo Studio, I also went in and experienced it. There are many resources available for developers, including technical courses. , cloud resources for running model training and verification, etc., and will also hold technical competitions from time to time. If you have energy, you can pay more attention to it, which is also an opportunity for technical combat. The community link is here, and you can experience it yourself when you have time: