The recent trial staged by Nokia and Japanese Telco KDDI claiming to be the first application of LTE-Broadcast (LTE-B) to vehicle-to-vehicle (V2V) communications represents another small step towards widespread autonomous driving. The test was carried out to test whether broadcasts to all cars within proximity would work as intended for V2V applications.
But it also illuminates how much remains to be done to build the required underlying infrastructure. There is a big difference between putting a few self-driving cars on the road and making the system work safely on a large scale where autonomous vehicles must coexist with each other but still cope with many legacy vehicles on the road. This will require full V2X communications where vehicles communicate in real time with each other, roadside systems and centralized applications (Vehicle-to-Everything).
On these fronts the use of LTE-B (one-to-many) rather than unicast (one-to-one) for low latency V2V connectivity represents progress by demonstrating a more efficient way of distributing information from sensor networks to all vehicles in the vicinity, minimizing bandwidth consumption. It is notable that LTE-B is gaining more traction for such telematics applications than distributing video as originally conceived. It allows specification of hotspots which could be stadia receiving video relating to an event taking place inside but in this case are clusters of sensors transmitting simultaneously to multiple vehicles while in the vicinity.
Nokia’s MEC (Multi-access Edge Computing) platform provided the core of the trial, keeping constant communication with the cars, forwarding location, speed and directional data to roadside sensors. Then, in an emergency, a driver, or eventually a vehicle itself in autonomous mode, can inform other vehicles for immediate action to be taken. This then was Vehicle-to-Network-to-Vehicle (V2N2V) communication, the first of two use cases demonstrated in the Nokia/KDDI trial.
The second use case was to investigate automated in-vehicle navigation through so called Network Real-Time Kinematic. This also addresses safety in autonomous or even assisted driving, exploiting existing satellite based and other geo-location systems to distribute real time positioning information down to cm level accuracy. Therefore, very low latency is critical for otherwise the information would be out of date and therefore by definition not high precision. The objective here was to investigate potential for cooperative driving rather than individual vehicle decision making, introducing collision prevention as a first layer of safety with the sensor base systems then kicking in for collision avoidance if this fails.
Such trials also indicate the challenges still to be overcome to bring on ubiquitous autonomous driving. One of them is dealing with exception conditions such as a car that is out of control and not signaling its position, driving in thick fog, or when unable to access cellular signals. As in aviation ‘fly by wire’ redundancy must be built into the system at multiple levels so that there is always some way of communicating with enough sensors to determine position accurately and be able to respond to nearby vehicles or events in time to take evasive action when necessary.
There is a general need to have some visibility beyond line of sight, especially on small winding roads where there is a risk of encountering oncoming vehicles with little time to pull in, especially if they are driving too fast. The remedy is effective non-line of sight imaging (NLOS) and recently a team at Stanford University in the US published details of algorithms capable of reconstructing images just out of sight by measuring time taken for reflected rays of light bounced around a corner on objects to arrive back at the emitter, which could be located in a car. The principle is that although these rays take different times depending on the exact path they take the average can be calculated and used to determine the position of each point on the surface of an object. The image can then be reconstructed accurately from this statistical analysis.
This is a highly computationally intensive statistical technique which is not feasible at present for implementation in vehicles travelling at normal speeds. But there is scope for optimization and some such technique will greatly enhance the safety of autonomous or for that matter manual driving, given that we have no way to see round corners at present other than by spotting the headlights of approaching cars at night.
There is also a mobile industry angle to trials such as this one in Japan, which is that by demonstrating the efficacy of 4G for an important aspect of autonomous driving Nokia could be said to be undermining the case for 5G, which it has been cheering loudly for. After all autonomous driving in general has been promoted as a major use case justifying the big investment that 5G will require. There is no doubt that 5G would be superior to 4G but the question is whether or when the gain will justify the investment. Trials such as this one may help prolong the life of 4G and postpone migration to 5G.