Why Automotive Autonomy Needs a Network Upgrade
August 07, 2024
Blog
![Why Automotive Autonomy Needs a Network Upgrade](https://data.embeddedcomputing.com/uploads/resize/320/200/external/data.embeddedcomputing.com/uploads/articles/primary_images/1723050205.jpg)
Autonomous vehicles have seen a major backlash in 2024, with high-profile accidents sparking public ire and forcing automotive companies to reassess their autonomous vehicle rollout strategies. Vehicular automation is a process still in its infancy, and the road from semi-autonomous vehicles with advanced driver assistance (ADAS) features to fully autonomous driving (AD) will be paved by next-generation in-vehicle networks.
Fittingly, OEMs are now demanding a network architected on day one of a platform’s release that can meet the compute demands of ADAS features available today while also being able to scale to the future demands of fully autonomous vehicles.
Automotive autonomy is categorized into five ascending levels of ADAS complexity, with level one including basic driver assistance features and level five achieving full AD. As much as any vehicle runs on gasoline or electricity, a vehicle at any level of autonomy also runs on data—massive amounts of data that must be processed extremely quickly to determine when to move, how to move, how quickly to move or to stop moving in a safe position in the road.
To make such data collection possible, autonomous vehicles rely on multiple sensors at the edge of a vehicle’s network and physically located outside of the vehicle cabin—all of which equate to an exponential increase in bandwidth in the vehicle.
Sonar sensors for near-object detection are at the low end of data generation, by bandwidth. Radar determines how quickly objects are moving in relation to the vehicle. LIDAR sensors are often large and placed atop a vehicle, using a pixel array of reflected light points to build a 3D map of the vehicle’s surroundings.
Supplementing these sensors are cameras, up to about eight inside vehicles on the road today with some OEMS predicting up to 30 cameras being needed in future autonomous vehicles. The cameras’ bandwidth demand increases further as they evolve to 4K resolution, with dynamic range and higher bit counts allocated to color representation.
So, a question emerges: What does an in-vehicle network look like that can move all that data effectively throughout the vehicle, from creation at the edge to aggregation at the vehicle’s brain, the CPU?
It does not look like the network found in the typical vehicle on the road today, which relies on a complex domain-based networking infrastructure with silo-ed ECUs across vehicle functions, from infotainment and lighting controls to ADAS.
Instead, building and implementing a centralized architecture will streamline the aggregation of collected data into zonally located ECUs. Building that centralized architecture through sufficiently advanced automotive Ethernet standards, supplemented by leading-edge chips at the physical layer and switching higher up the protocol layers, equates to significantly faster data processing. The sensor data previously highlighted can be aggregated at a point near the sensors and can be further aggregated to a vehicle’s backbone, which is typically provisioned to be a much higher bandwidth than that needed at the edge.
The typical passenger vehicle on the road today is running on a 100 Mbs or 1 Gbps backbone, but solutions are coming to market soon running at 2.5Gbs, 5Gbs up to 10 Gbps—cutting latency 10x, meaning a lot more data can be moved much quicker, giving autonomous ECUs the time they need to make the necessary decisions on a multitude of sensor data that keep passengers and pedestrians alike safe.
Once these networks enabling high-speed data processing and ultra-low latency can be implemented into new-vehicle production, the road to full autonomy will become much clearer, much safer and much more reliable.