When the world is treading on the driverless car path, it is important to understand whether the driverless vehicle is really intended for every terrain, or to put it in a better way, can driverless cars handle every terrain? For instance, if sensordthe driverless car is in a hilly terrain it will not be able to process information because the availability of information will be limited. Think of something like a hairpin turn, you cannot see what is coming, is there a technology that sees through mass, and lets you know what is there on the other side? No technology using light waves (anything in the electromagnetic spectrum) or sound waves at least can help the autonomous car with this.
The HD maps will be able to solve the problem of the static environment of the cars but the real problem is dealing with the dynamic environment. The one way this problem can be dealt with is using satellite images of the area, which as of today is not permitted since it is a security concern.
The terrain being talked about here is like the one shown in Figures 1 and 2. In such terrains where the cars are not at the same level, the technologies based on LoS become dysfunctional. Only after traversing a few metres when these cars come face to face these technologies work. Imagine a situation like the one shown in Figure 2. The two cars shown here are likely to collide at the turn. The tricky part here really is the reaction time of the passive safety systems? Within how many meters can they alert the driver? For something like this on a hairpin turn the car needs to respond within a couple of meters. Are autopilot systems capable enough for this? Otherwise these cars are not made for driving on all kinds of terrains.
The situation shown in Figure 2 would hold with 2D RADAR systems. The problem can be solved using a 3D RADAR system or the LIDAR systems that autonomous cars are already equipped with. The LIDAR systems span the whole vertical space as well along with a 360 degree view. Refer to Figure 3 for the details on the same. The rotation of the rotating mirror shown in the figure causes a spanning of the vertical space. With this LIDAR system atop the cars shown in Figure 2 the problem of detection of the cars is gone. This kind of scan of the whole 3D surroundings of a point is also done by 3D RADARs which thus becomes an alternate to LIDARs in autonomous cars. The working of 3D RADAR is as shown in Figure 4.
The one problem for which the right technology is still needed in an autonomous car though, is one which does not depend on Line of sight communication and has high speed. Think of the terrain shown in Figure 4. When one car is going on this road and another car is just near this steep turn on the other side, hidden in view by the mountain there is no technology in autonomous cars which will help it take the right decision in this terrain in the right time. By the time the vehicles on this turn actually “see” each other the distance between them will be less than a metre, too small for the car to take a meaningful action.
The one way in which one can see the problem of terrain dependence totally solved is with the connected cars technology. Each connected car is connected to the cloud and so each car connected car can actually see any other using the GPS locations so shared. But for this technology to work, every other car needs to be a connected car. So each autonomous car needs to be a connected car as well. And this situation will be further simplified if after having located each other the cars talk to each other as well, that is, V2V (vehicle to vehicle) communication would make autonomous cars more effective and their artificial intelligence more human-like.
When it comes to terrain not just the geography of the place but also the kind of surface should be a matter of concern for the autonomous car driving to become a better experience. Jaguar Land Rover has been working on this, motion of cars on various kinds of surfaces. Not just Jaguar Land Rover but even Ford has tested its semi-automated cars on snow.
Complete automation in automotive seems not too close today. The variables that an autonomous cars face, the static environment, the dynamic environment, the decisions that need to be made using these variables can be made less realistic. When an autonomous car moves on the same route repeatedly, the static environment remains the same, and so the computations become easier. V2V communications is another way of simplifying computations for autonomous cars, by making the dynamic environment easily understood. They make cars talk to each other, as cars with human drivers would talk or exchange gestures while making critical decisions on the road.