In recent years there has been extensive efforts to produce automated vehicles ranging from highway automation in the form of traffic jam assistants and road trains (platoons), to urban public transport and valet parking. The main two trends are individual automation and cooperative automation. What is the difference?
Well individual automation relies on its own extensive vehicle sensors to travel while cooperative systems rely on their own sensors but also on communication systems with other road users and/or the infrastructure to perform their task.
An example of cooperative systems would be platooning, similar to what you see trucks today on highways doing. Volvo recently publicised the road train system which utilises V2V very similar to TNO’s C-ACC. This communication allows vehicles to travel at very short distances from each other and at fairly high speeds since the vehicles act essentially as one. However that also means that the leader currently is a manual driven vehicle. I wonder who will choose that spot!
Similarly TNO recently presented their CAEB, a system that connects VRU’s (i.e. Cyclists, pedestrians) through V2x to cars and enables them to brake before the on-board sensors even see them.
The US NHTSA also started recently pushing the requirements that all new vehicles to be equipped with V2x technologies.
Its well accepted that if every vehicle (i.e. car, bus, train etc) and every person were connected and could communicate their location and intent things would be quite different.
No accidents, no traffic jams, perfect on demand public transport etc etc etc….
But what is the reality of this?
Well let’s take a look at Highways for example, it’s quite obvious that trucks travel in convoys and usually on the right most lane but how many private vehicles travel in convoys? It’s quite difficult to travel in a convoy if we all have different destinations, some of us are in a rush, some not or just simply a basic human individualism. Either way, if we consider the effort to equip each car with fast and safe V2V, assuming for example 400k (rough numbers) new cars per year in a population of 16M, assuming everyone has a car or will buy one, it would take around 40 years to achieve this. So that s not a fast way to do it.
There has been also much talk about the security of these V2x systems which is still an ongoing issue. How do we protect such a system.
What will the effects be of only a few cars being able to communicate with each other or someone forgets to switch on their transceiver? The truth is nobody knows and that’s something we need to understand.
Well, everyone pretty much today has a mobile phone so isn’t that the basis? Yes, but how many people want to transmit their current location even if anonymous or even their destination?…not many I imagine, and who pays for it? How?
So the question is what do we do in the mean time?
Well this is the reason why most vehicle manufacturers are looking into individual automation, cars that can perform autonomous driving tasks independently. Several Car manufacturers have recently announced their new autopilots for highways and traffic jam assistants .
A major problem we face is that we trying to sense a 3d world with mostly 2d technology. The most widely used technologies are radars and vision systems (camera based). There are still a lot of issues with these sensors that make it difficult to enable 3d data. Let’s take Stereo vision, how dynamic is it really? Currently its quite a fixed system and 3d data is given only in a specified region of its FoV.
The biggest problem is computational load which roughly translates to increased costs. Nobody is going to pay thousands of euro on top of the standard price of their preferred car for limited automation and hence the limited functionality offered by high end cars only, equipped with technology today.
I have heard many people argue that laser scanners are the future due to their accuracy of the raw data. I would possibly agree if it wasn’t for their price tag and the post processing from 3d-point cloud to targets.
Take for example the 70k euro laser scanner on top of the google car. You could argue well is it necessary?
At the TUD we have taken a step back first defining the sensing metrics of each environment independent of any sensor (i.e. FoV, Range, Resolution, accuracy etc) and used these to define sensor sets. The work is ongoing.
There is a rather big issue with regards to accurate digital maps not mention how do you deal with dynamic issues like road works. This is also the primary use of a such a sensor.
Well a lot of effort has been put into accurate digital maps and e-horizon systems to provide automated vehicles with required map data. These efforts continue also at TUD.
Let’s not forget that all sensors around a vehicle are limited to line of sight which in the real world translates into having to respond to last minute situations.
The next stage is the decision making, how and when to make decisions that are related to safety and/or comfort. For example, let assume overtaking on a highway, not a big issue for most but how do we train a machine to take a chance for example? We humans can estimate when a gap presents its self and we make a decision to go for it. How does a machine take a chance that would most likely violate its safety instructions? The difficulty level increases in Urban environments were busy two way narrow traffic streets exist. The likely hood is that you will be stuck behind a crawling vehicle or a delivery truck.
And what happens if the driver decides to intervene ?
Should we wait until V2x systems become widely available to attempt any such manoeuvres?
Unfortunately many of our decisions are going to be have to be made on the basis of incomplete, dubious (hacked) or limited information (line of sight sensing) and must be able to deal with the unexpected daily life throws at us. …or we can just wait for a few years, maybe!
The story continues on the next edition of this blog!
If you are triggered by a possibility or want to be part of this challenge…get in touch with us!