the AV Ecosystem Explained
Read the Article and Download the Infographic
Automated vehicle systems are a complex endeavor for companies that are trying to carve out a slice of the ever-expanding AV eco-system. There are literally hundreds of companies and thousands of products competing for a slice of the technology pie necessary to support automated vehicles.
VSI’s Eco-System Infographic shows leading companies that offer components, systems, and tools necessary to complete an AV build. The chart is partitioned into the various technology domains and includes low level components as well as subsystems, development platforms and complete vehicle builds. What this chart does not include are the cloud assets necessary to support automated vehicle systems, particularly L4/L5 deployments that rely on cloud resources and data center assets.
The level of integration is also represented in this chart. The bottom tier are development tools and apply throughout the development process. The vertical columns represent the various functional domains that collectively apply to a fully automated vehicle technology stack.
The horizonal layers near the top represent AV platforms of one form or another. The first horizonal layer represents AV Development Platforms (aka. AV domain control). The middle horizonal layer are fleets and various mobility services and/or platforms and represent known development vehicle projects. Some are operational, but many are still in concept. The top layer are largely OEMs who are developing automated vehicle systems either for series production (L2/L3) or for full automation (L4/L5).
Download the Infographic
Fill out your name, company and email address. After clicking 'submit,' a message will display with a link to download.
Development tools are vital for designing sophisticated AV systems. Within this category you have several tools necessary for modeling the performance requirements of the system long before anything gets built.
For the development of the systems there are many companies that offer simulation tools for testing the performance of the systems in a virtual environment. These tools are used for testing the performance against a virtual environment where scenes, actors, sensors, and physics can be modeled. Some of the simulations offer the ability to test the individual components while others are used to test the performance of algorithms.
Within the development stack you have lots of tools for the development of runtime components. Within this category you have compilers and code checkers that look for errors or anomalies in the code base and optimize it for deployment against a target hardware stack. Additionally, there are many signal management tools that are used for timing and synchronization of processes from sensor fusion, to ECU to ECU communications.
The processing stack is a busy space because the different functional domains within the AV stack have different processor requirements. Many of the processors offered by the companies in this space include specialized processor logic that is optimized for parallelism which can process many streams of data in parallel. Other processor architectures are specialized accelerators for graphics processing, image recognition or artificial intelligence. Once again, the common thread (no pun intended) is parallelism in which massive amounts of data and instructions can be processed simultaneously.
Some of the processors represented in this domain are SoCs that contain multiple processor instruction sets on one chip. Some include DSP technologies, or FPGA fabric from which new instruction sets can be dynamically changed or altered. Many of the SoCs contain a host processor and often this is based on ARM cores. ARM cores are RISC processors designed for the host functions of an SoC.
There are processors in this space that are designed for functional safety or safety critical tasks and may include dual core lockstep processors that are required for ISO 26262.
Some of the companies represented in the processor space don’t even sell chips but rather license their instruction sets that can be hosted within an FPGA or may be burned onto a host chip.
Not surprising the sensor stack is the most crowded of them all. Here you have a continuous stream of new entrants many of which are developing new and/or lower cost sensors or sensor modules. For example, Lidar is the most diverse space due to the different approaches as well as new technologies such as solid-state componentry.
Up until this point Lidar has been prohibitory expensive except for commercial fleets where you can afford a more expensive sensor package.
Lidar is basically 3D imaging if the resolution is dense enough via a point cloud, but the limitations arise for long range applications. New solid-state devices employ beam steering rather than using a rotating drum. Low cost devices flash Lidar devices are also coming which are basically limited resolution that can identify the presence of an object but cannot classify or produce a point cloud.
Meanwhile, radar is still widely relevant and even gaining in this space because of new developments. Here you have a handful of small form factor millimeter wave radar units that have flexible antenna configurations that can yield more granularity than previous radar. You can also achieve 3D measurements if you employ multiple antennas at different places in the car.
Camera sensors are widely represented by the companies in this domain. In the case of camera, you have many companies that make imagers -- the sensor chip that is uses a raw component in a sensor package whether it be a module or a board level product that couples sensing with some processing. Image processing requires an exhaustive pipeline of processing power and new image sensors can handle some of the preprocessing requirements while still sending raw data to a central architecture.
This space is also represented by companies that offer sensor fusion applications whether it be hardware or software. But most sensor fusion applications still fuse object data and not raw data. Further, in many cases, smart sensors produce object data within the sensors, while other sensors send raw data to the main processor — where objects are produced before it is ingested into the fusion engine. For AI applications there are different schools of thought when it comes to object versus raw data.
Data Connectivity Stack
The data connectivity stack is a combination of hardware and software solutions that support the movement of data along the in-vehicle networks or via wireless networks outside the vehicle. Here you have various tier-ones that make connectivity modules (ECUs) that can handle the data traffic, compress/decompress or encrypt messages where needed. Others in the space produce network interfaces and switches that may be a component within the connectivity space.
Meanwhile, wireless modem makers are a vital member of the data connectivity stack as future AVs must maintain connections to service providers and data centers because they are always talking to the cloud.
The mapping stack is a combination of various localization assets necessary improve the performance and safety of automated vehicles. Precision map data enables the vehicle to understand where it should be without having to rely exclusively on your environmental sensors. For example, in most automated applications image sensors track lane markings. And while lane markings are the most common way to determine where a vehicle is supposed to be, foul weather will prevent AVs from operating properly if relying exclusively on this.
Precision map data will increasingly gain in importance for all levels of automation because it will heighten the performance and safety of these technologies. Furthermore, higher levels of automation (L4/L5) will require maps to properly perform grid fusion that will be necessary in dense urban settings.
Maintaining up-to-data map data is forever a challenge for the mapping companies so new technologies are emerging that allow fleets of vehicles to pick up on changes to the road infrastructure. As a result, there are companies emerging that allow the aggregation of real time road conditions so that this information can be relayed to oncoming vehicles. Also known as crowd sourcing, this method will be necessary to support the dynamic nature of roadways where road conditions, closures, detours, and transient conditions are always changing.
The algorithm stack is a collection of software assets used throughout the AV processing pipeline. Many of the companies in this space offer algorithms for the perception side of the equation and here the task is further divided by the different sensors that are used.
Additionally, the process of modeling the environment in real time requires advanced perception algorithms which may apply neural networks. Behind the scenes in AI are sophisticated training methods that create new inference models which are essentially AI-based algorithms that get redeployed to the AVs.
AI is one of the hottest areas within the field of automation because AI-based algorithms have the capacity to reason with partial information. Furthermore, the collection of data from fleets of vehicles enable the algorithms to improve with time as fleets are constantly collecting information on new scenes that may not have been thought of in previous training exercises.
Within the algorithm space there lots of new developments with respect to the full software stack. Several companies are offering stacks that support the whole processing pipeline. These tightly integrated software stacks relieve automakers from taking on the tough task of developing their own algorithms.
Lastly you have algorithms for the control side of things. Often overlooked control algorithms must by dynamic to account for different physical requirements of the vehicle. While control algorithms for ADAS tend to be deterministic in nature control algorithms for highly automated vehicles will need to adjust to themselves to account for changing conditions or performance requirements, especially for L3+ vehicles where a driver is out of the loop.
Safety and Security
Within the context of safety and security there are a diverse set of requirements that are necessary to either secure the safe performance of the vehicle, as well as protect the vehicle from malware and outside security threats.
On the safety side you have Functional Safety which is a process from which AV technologies are designed and developed. Functional safety is critical for automotive best practices and many of the companies in this space offer safety rated components (either hardware or software) that are designed to minimize malfunctions, spot abnormal behavior and even instruct a safe failure. Many of these technologies are applied to the runtime components deep within the software stack (such as RTOS).
Meanwhile, within the context of cybersecurity, you have a lot of development that assures protection from malware and other cyber threats. While measures for cyber security are applied at many places throughout the hardware and software stacks, there are sophisticated cyber protection layers that are dedicated to the task of identifying and shutting down unauthorized code or instructions.
VSI conducts applied research on automated vehicle systems. Our research solutions are designed to help companies make better decisions when it comes to the design, development or planning of automated vehicle strategy. VSI’s customers are global and represent leading OEMs, tier one suppliers, components makers and leading technology companies.
VSI’s AV Eco-system research decomposes the entire value chain with assessments on companies, products, and technologies. Furthermore, VSI provides a continuous stream of technology insight to help you tackle tough technology decisions.
VSI’s Pro research is designed for companies that are building AV solutions and need technical research and support of complex AV functional requirements such as sensors and sensor fusion, domain control, AI, decision, control and safety. Furthermore, VSI conducts functional validation of AV components to help prove the improvement in safety or performance that accompany a solution.