Autonomous driving will be the future of driving and one of the most important future revenue streams for automakers. A faster industrywide shift to the era of autonomous driving will result in significant improvements such as less traffic congestion and pollution as well as fewer accidents. However, the main barrier is that the technology is not yet mature enough to take full control of the vehicle.
Autonomous vehicles are part of the larger paradigm shift toward software-defined vehicles, requiring major software investments from automakers. If crucial software and architecture development decisions go a bit sideways, it can have long-lasting effects.
We are currently only seeing the first steps toward Level 3 autonomy. Mercedes-Benz recently introduced its Level 3 Drive Pilot system, which is first available in Germany only.
The amount of different sensors needed -- and consequently the data needed to be processed in real-time -- means software development costs, chip performance levels and associated energy consumption rises radically between different levels of automation, for example, from Level 2 to Level 3. The increase is even more substantial going from Level 3 to Level 4.
Reducing the complexity of architectures, chips and software is one of the central challenges to reaching autonomous driving faster. For example, the industry is in transition from a distributed electronic controller unit (ECU) architecture into a consolidated ECU architecture.
Autonomous driving requires significant computing power, and power consumption is critical, especially in electric vehicles.
Based on what we see, the current chip generation is probably not able to fully deliver what complete autonomous driving requires. A new generation of chips is needed to address the power consumption and safety certifiability of the system. And software that’s able to utilize the underlying processors in an optimal way also leads to substantial power consumption savings.
Autonomous driving will create a crazy amount of data -- more than 10 gigabits per second -- that needs to be analyzed in real-time. The biggest contributors to this data are cameras and lidars. We estimate that a single camera with 1280 by 720 pixels resolution that records at 30 frames per second in full color will generate 450 Mbits/s -- and these cars will most likely have at least three of those.
A lidar with a 5 million point resolution running at the same 30 frames per second will generate around seven Gbits/s. The other sensors, such as radar, sonar and GPS, only generate a few hundred Kbits/s.
Cars now have as many as 100 processors, which is very power-consuming, and it adds slowness to the system as well due to the need to transfer data around.
The low memory bandwidth of embedded systems leads to slow and energy-consuming intercommunication between processors. This complexity can result in very high power consumption in autonomous driving for electric vehicles and therefore, drastically reduce range.
Instead of just waiting for the next chip generation to possibly solve this complexity, automakers should move to a much more consolidated architecture with the current chip generation. With smart software, they can already achieve a lot today.
Autonomous driving will reduce the number of cars. Shared mobility will become normal, and those cars that would otherwise be idle for most of the time are available for better use.
A world with fewer cars would have positive effects. For example, parking, roads and other parts of the infrastructure require for vehicles account for about 20 percent of a city such as San Francisco. If even half of that infrastructure can be repurposed toward greener options, such as park, then the entire plant benefits.
Cooperation between companies and ecosystems is crucial. The automotive industry is such a large field, and the issues around autonomous driving are so complex that no company can solve autonomous driving challenges alone.
