Get best of the week

Autonomous cars: a stack of technologies

image

The industry of unmanned vehicles is at a level of maturity comparable to the traditional automobile industry 100 years ago. Henry Ford produced at the Model T factory not only the car itself, but also the wheels, as well as most other components and parts of its own production. One hundred years later, several large and hundreds of small suppliers provide 70% of the components of a typical automobile vehicle, while car manufacturers make only 30% of their own.

The total revenue of only the 10 largest suppliers grew to $ 315 billion in 2017, and the revenue of the 100 largest suppliers around the world approached $ 800 billion, according to Automotive News. And it makes sense: the details under the hood usually do not allow the vehicle manufacturer to distinguish it from a competitor, or, in other words, the end user - that is, the owner / operator - usually does not care or, as a rule, does not even notice if there was a radar sensor for An adaptive cruise control system made by Bosch, Denso or someone else while it is reliable and working.

The trend in unmanned vehicles continues to go in the opposite direction. For example, Waymo designed and is building its own lidar, Cruise bought a lidar company in 2017, and more recently, Aurora also acquired a lidar company. The reasons are simple. Almost everyone (with the possible exception of one person) believes that lidar is crucial for developing the safest and most reliable unmanned driving system. In other words, companies think that they will have a competitive advantage if they, in their opinion, are key in this area. On the other hand, this perceived short-term advantage is very vague - there are 70 such companies (not including Chinese), and it is not clear which technologies will prevail in a couple of years. And again, being a passenger of a robo taxi in a couple of years,I am not worried about the manufacturer of the individual components if the car provides a safe and comfortable ride to the destination.

In the medium and long term (mainly financed by venture capital funds), companies with a full stack (i.e., those trying to maximize the implementation of all components - both hardware and software - within the company) operating in the field of unmanned vehicles will notice that costs and complexity will greatly increase.

The founder of a (relatively small) full-stack company recently told me that they "can afford to create a full stack on their own because their addressable market will be very large." I am not sure that he fully understands the economy of this scale. A large and profitable market attracts more competitors, ultimately lowering prices. A supplier supplying products to 10 manufacturers can obviously offer a lower cost. And again, this is precisely what led to the fact that car suppliers contributed more than 70% of the cost of a conventional vehicle.

Recently, partnerships have been developing in the field of unmanned driving, which were not previously visible. Volkswagen refused the contract with Aurora and instead invested in Argo, and even presented Argo with its unmanned automobile subsidiary AID as part of the deal. BMW and Daimler are joining their drone development divisions, and it is rumored that Audi will join too.

We think this is just an intermediate step. Ultimately, the unmanned driving industry will see the same transition. I call it the decomposition of an unmanned car stack. The entire stack is too large, too complex, too expensive, too resource intensive for most companies to develop on their own. This includes too many different disciplines and skills.

The unmanned car stack consists of five main groups: hardware, external software and data, firmware, various methodologies that together lead to product development.

1. Hardware stack


image

The hardware stack consists of a vehicle platform, often customizable or customizable for a particular application, which contains interfaces to actuators, i.e., transmission, brake and steering systems, as well as electronics. In addition, the components include the on-board computer (s), the communication system inside the vehicle, as well as the cloud, as well as data recording and storage components. Sensors include GNSS, motion sensors, lidar, radar, camera, and sometimes ultrasound.

2. External software


image

External software and data includes maps (with different layers, see, for example, the Lancelet 2 document for details). Maps should be created, annotated with meta-information, updated and distributed - also in parts - while maintaining consistency across the entire map database. Highly autonomous vehicles are often operated as part of a fleet, which requires fleet management, fleet routing, teleoperation, self - and remote diagnostics for unmanned vehicles. The sensors of one autonomous vehicle generate up to 64 GB / s (or 8 GB / s, or 480 GB / min, or 28 TB / hour, or 560 TB / day, or 200 Petab / year). The fleet obviously creates a multiple of this. This amount of data must be recorded, saved, annotated,analyzed and manageable. Software developers need a software development environment that delivers productivity. Tools include data reproduction, data visualization and the ability to model data at various levels.

3. Methodologies


image

In the development process, various methodologies are used, including system design, software architecture design, hardware design, interaction design. Tests should be developed at all levels, including program unit, regression, integration, SIL, HIL, vehicle tests. Other aspects include functional safety, rules, homologation, security, protection, verification and validation.

4. Firmware


image

The built-in software stack consists of an operating system (I hope, in real time for critical security systems, not just Linux), which itself consists of a kernel, a scheduler and drivers. On top of the OS is (at least in a well-designed system) a software structure that abstracts from the many aforementioned components, such as the OS, computer hardware, touch interfaces, data recording, playback, visualization, and middleware. It should provide support for security, protection and diagnostics. ROS, the robot’s operating system, is de facto the standard framework, and the ROS 2 collection of articles on design provides a more detailed overview of the components included in the robotic framework. Here at Apex.AI, we developed a commercial,soon certified security fork of ROS 2, which we call Apex.OS. At the top of the framework are algorithmic components. Perception refers to the processing of information from sensors into a brief environmental model. Localization is the location of the vehicle relative to the lane, road and world shown on the maps. Understanding the scene penetrates the semantic understanding of the perceived world. Driving decisions are made based on a number of goals and depending on the environment, the desired vehicle movement is planned and sent to the vehicle's actuators via the controller. Many of these algorithmic components are implemented using modern methods of artificial intelligence (AI), which can achieve accuracy similar to human,but make new demands on many parts of the stack.

5. Product


image

All components must be integrated into the product, which here refers to the application that interacts with the user / operator. The application is adapted for its intended use, configured for the capabilities of the stack that supports it, and released. System integration for highly autonomous work is a collaboration with several players, and with a well-designed stack, it can quickly go through the definition, configuration, testing and release.

The enabler for a modular and expanded stack is a common architecture. ROS provided us with a standardized and open architecture and open source implementation managed by the foundation. We expanded the ROS model to the application stack and recently became one of the founders of the Autoware Foundation. The Autoware Foundation develops a functional architecture for self-driving and builds this architecture and a fully open source reference implementation. More than 35 companies and organizations have already joined the Autoware Foundation. Join this powerful group to help create the standard.

Waymo would not be ahead of everyone else if they were waiting for the development of the stack decomposition, but this is the pioneer burden. Everyone else who tries to catch up will do it faster and cheaper, supporting the decomposition of the stack, helping to set standards and choosing the right partners, in the hope that it will not take 100 years to reach the next level of maturity in the autonomous industry.


image

About ITELMA
- automotive . 2500 , 650 .

, , . ( 30, ), -, -, - (DSP-) .

, . , , , . , automotive. , , .

Read more useful articles:

More articles:


All Articles