Don’t Ask Faction if You Can Remote Drive Vehicles Over Cellular Networks, Ask Your Telecom
You cannot scale a business that remotely drives a fleet of vehicles given the current wireless infrastructure. Let’s explore why.
To scale autonomous vehicle fleets today, a key need is to have a human in the loop to supervise select operations. The best autonomous systems can drive on the freeway approximately 95% of the time and approximately 85% of the time on surface streets without human intervention. Thus, there is a non-negligible amount of time when autonomous cars need help in a timely fashion, forcing regular supervision.
For the last several years it was believed that AI would be able to drive vehicles 100% of the time within a matter of months, so the industry’s best practice was to have a safety driver in the car for those edge cases where the AI could not drive. In recent years, some of the big players have moved safety drivers from the vehicle into an office, seated at a teleoperation workstation, where they remotely manipulate steering, braking and an accelerator pedal.
A recent trend takes this approach to the extreme and advocates 100% teleoperation. That is, have a human driver control a vehicle remotely at all times.
Ain McKendrick, Faction’s CEO, covered why partial to 100% teleoperation is a bad idea here.
There is more. While technically feasible, autonomy with teleoperation or 100% teleoperation, will not scale if you’re operating a business. A different paradigm is needed to operate fleets of driverless vehicles that account for the wireless infrastructure capacity.
For the sake of simplicity, let’s only discuss the use of cameras to support a solution for remotely assisting a fleet of driverless vehicles and ignore other data, including sensor data, moving in/out of the vehicle.
The data from the cameras is required to travel over wireless networks to the nearest cell tower and make its way to a cloud server. This poses the following problems:
- The amount of data that needs to move upstream per vehicle;
- The amount of data that is going to the cloud once it enters the network;
- And finally, the amount of data hitting cell towers at once in high-density urban areas.
While you can throw money to solve for points 1 and 2, and buy more bandwidth from your service provider, you can not resolve the issue for 3.
Note that, for the sake of simplicity, in the calculations below I am also ignoring the fact that the wireless infrastructure is asymmetric and upload capacity is smaller than download capacity.
Additionally, there are physical limits to what wireless networks can carry today and limits to what the carrier can sell you. The only way to get more capacity is to build your own infrastructure and maintain your own cell towers, assuming you can secure the radio spectrum needed, or wait for full, true 5G penetration. Arguably the latter is several years out.
Will 5G solve (3)? To some extent. The 5G spectrum could be priority sliced to enable driverless fleets to operate in this fashion albeit, not to give away the ending, next you run into latency constraints. Fundamentally we need to rethink how we remotely assist a driverless vehicle. Going back to today, the reality is that anyone wanting to run a driverless operation should assume 4G capabilities at best. So if you want to run a business, you need to plan and engineer your solution within 4G, with the ability to grow and leverage 5G when and where possible.
How Bad of a Problem is this?
Assume a density of vehicles per square mile of 3,800, which happens to be the density of vehicles in Los Angeles, and say that 25% of them will soon be driverless. Imagine these driverless vehicles are within the same square mile and they all hit the same 4G cell towers. Now, compare how these driverless vehicles interact with the towers if they were Faction-enabled vehicles using TeleAssist®, a best of breed Autonomous vehicle, or are 100% teleoperated. Autonomous vehicles and 100% teleoperated vehicles use remote driver monitoring or active control of the vehicle at all times. This requires low latency, high bandwidth connectivity to stream the cameras, and to send steering, braking, and acceleration commands back to the vehicle in near-real time.
Faction’s TeleAssist® uses a different paradigm: it provides remote assistance on-demand, without remote driving and low demand for bandwidth or latency. It uses 720p equivalent cameras, which only stream full resolution during burst time, or about 15% of the time in urban environments. Today, Faction vehicles with TeleAssist® utilize an average of 32 Mbits/sec per vehicle and we expect to drop to 18.5 Mbits/sec soon. 100% teleoperation approaches to driverless vehicles need 16 Mbits/sec for each high resolution camera (we are assuming 1440p equivalent at minimum), multiplied by four (4), totaling 256 Mbits/sec per vehicle.
The table below explains what percentage of the total cell tower’s capacity would be used by a Faction fleet vehicle vs. a 100% teleoperated fleet. To compare apples to apples, we are assuming that each vehicle uses 4 cameras (front, left, right and rear), albeit they are equipped with more.
Let’s review: if you were to deploy a fleet of 950 Faction vehicles in urban Los Angeles, the local cell tower would need to dedicate 76% of its capacity to said fleet. To be fair, vehicles located in the same area will likely hit a few towers (about four); statistically, we would be looking at 19% of each tower being used for TeleAssist®, which is a sizable, but reasonable number.
More to the point, 19% is physically possible.
However, if a fleet of the same size of 100% teleoperated vehicles were in the same location, it would require 3,595% of the tower’s capacity. Under the relaxed assumption that the fleet would balance over four towers, we are asking each tower to provide 898% capacity. Assuming you could absorb 100% of each 4G tower, you are 498% short!
It is not physically possible for the wireless infrastructure to support such business today.
That’s the problem.
If you want to run a teleoperation business, don’t ask Faction why you can’t remotely drive a fleet of vehicles, ask your telecom.
—
Andrea Mariotti is VP of Vehicle Engineering at Faction which develops driverless solutions based on light electric vehicles. The company believes the future of sustainable transportation is to develop driverless vehicles that are safe, cost-effective, and right-sized to serve a range of use cases for both business and passenger transportation needs. For more information visit www.faction.us
