Given this inevitability, the obvious question is: “Are we ready?”
As you might imagine, the safe and secure operation of millions of 2-ton projectiles simultaneously speeding freely along our roadways presents some imposing technological obstacles. Autonomous driving (AD) vehicles require an incredibly intricate yet reliable infrastructure. There is no room for error. Precise operation during every millisecond of drive time is critical to everything.
The compute requirements of an AD vehicle are mind-blowing. According to Intel, by 2020 each autonomous vehicle will generate a tsunami of data — more than 4,000 GB per day. That’s equal to the amount of data generated by 3,000 people, assuming the average internet user produces 1.5 GB per day. Following that math, a million AD vehicles will generate as much data as 3 billion people. And every bit and byte is dedicated to precision operations and safety — there can be no waste or excess.
Here is the breakdown behind Intel’s calculation:
- Video cameras ~20-40 MB/sec
- Radar ~10-100 KB/sec
- Sonar ~10-100KB/sec
- GPS ~50KB/sec
- Lidar ~10-70 MB/sec (laser-based light detection and ranging systems)
Of course, the huge amount of data generated by AD vehicles — coupled with the real-time service requirements — comes at the expense of network resources. Indeed, the workload of AD vehicles will require nothing less than the world’s most advanced network architecture.
A new communications ecosystem
AD’s inter- and intra-system communications necessitates the development of an entirely new technological ecosystem that brings together the cloud, the AD network infrastructure and the road infrastructure. Since seamless, real-time communications between these hierarchies is required, cloud transactions alone are nowhere near adequate for AD. That’s where fog computing and networking comes in.
The OpenFog Reference Architecture, published in February, provides a medium- to high-level view of system architectures for fog nodes and networks in IoT, 5G and AI — each of which is critical to AD. As shown in the diagram, the multi-tiered OpenFog architecture fills in the gaps along the cloud-to-thing continuum by supporting vehicle-to-cloud (V2C) and vehicle-to-everything (V2X) connectivity.
Sensor data and video/imaging content for AD vehicles create huge amounts of upstream traffic (from vehicles to cloud), as well as huge amounts of downstream traffic (from cloud to vehicles). This requires a distributed vehicle-to-cloud model that enables safety services. With fog, each vehicle has a distributed, onboard compute architecture with significant connectivity between multiple onboard compute nodes that support analytics, storage and other applications.
The fog network typically takes the form of fog nodes arranged in a hierarchy between the low-level control computers in the vehicle and the remote servers in the cloud. This hierarchy enables multiple AD services and helps determine which services need to use the cloud and which services are more efficiently conducted without the cloud, at the node and network level.
V2C technologies adhering to the OpenFog architecture enable AD processes while providing various services that assist the AD driving process (such as real-time, high-def maps). This can help vehicles on roads drive cooperatively with one another and to be aware of road hazards. Fog also includes higher-level mobile fog nodes in the vehicles that coordinate the functions of the lower-level processors that manage things such as powertrain control, sensing, collision avoidance, navigation, entertainment and so on.
Fog nodes are also prevalent in roadside units, so high-performance computation and large, reliable, secure storage is located within a short network hop of all vehicle positions. Above that, regional fog nodes coordinate the operation of roadside fog nodes, optimizing the smart highway for all drivers. And when the fog infrastructure communicates with the cloud, it is to ensure that everything is safe and efficient across the entire smart transportation system.
The fog architecture was designed to easily enable V2X communication and services. Each vehicle is a mobile fog node that communicates with the infrastructure, other vehicles, the cloud and outside entities, such as pedestrians and bikers.
Fog node functionality
Let’s look at four types of fog node implementations in the AD ecosystem:
- In-vehicle fog nodes provide distributed, onboard compute infrastructure for AD vehicles. The nodes process data from cameras, Lidar and other sensors. They communicate over a high-speed databus for onboard advanced analytics, path-planning and the rapid response needed for AD. Real-time visualization of hazards can be fed to the navigation system.
Additionally, in-vehicle fog nodes provide communications with other in-vehicle and infrastructure fog nodes for early access to travel safety-related information, such as road or lane closures, obstructions on the road, traffic incidents such as accidents, hazardous objects on the road, icy conditions and so forth. This enables AD vehicles to plan alternative routes dynamically.
- In-vehicle fog nodes, along the road and at the network edge, provide extended sensing capabilities such as smart cameras with computer vision in vehicles, smart cameras with computer vision in roads and roadside units with sensor capabilities for detecting metrics such as vehicle density and ramp length.
- Fog nodes co-located with road side units, such as traffic lights, streetlights or charging stations kiosks, provide local and regional services provision (for example, traffic information updates, accidents alerts, touristic guide information, shopping promotions).
- Fog nodes co-located with street cameras provide edge analytic capabilities closer to the source of video that can be vehicle cameras or street cameras.
Fog provides an entirely new set of network resources upon which AD applications can run their computation, networking and storage functions, located between the traditional resources in the car and in the cloud. The OpenFog architecture can greatly improve the performance, efficiency, bandwidth, reliability and feature richness of AD networks.
All IoT Agenda network contributors are responsible for the content and accuracy of their posts. Opinions are of the writers and do not necessarily convey the thoughts of IoT Agenda.