Alex Lidow, CEO of Efficient Power Conversion (EPC), assesses where the driverless vehicle sector is headed and the technologies which will help it realise its potential…

Autonomous vehicles have evolved much faster than any of us expected, fueled by the development of systems like LiDAR, but also deep learning tools that have digested the inputs from millions of miles of autonomous driving.

Today we have ‘workable’ autonomous cars from the likes of Waymo (formerly Google), Uber, and others. The difference is in the experience derived from all the autonomously driven miles and the diligence of these companies in learning from every one of those miles.

LiDAR adoption

LiDAR (Light detection and ranging) is used in the autonomous vehicle industry to create very fast, very precise 3D images of the vehicle’s surroundings. There are many ways to create a 3D image, but LiDAR is the most efficient and accurate overall. In essence, LiDAR units are the ‘super eyes’ of the driverless car. These ‘eyes’ can see farther, and understand exactly the location of everything surrounding the vehicle faster than any other technology and even a human.

An alternative method, used only by Tesla at this time, is to combine radar with cameras to ‘interpret’ the surroundings. This requires a great deal of computer power to interpret the various inputs accurately and quickly. Since it is an interpretation, as opposed to a direct observation in the case of LiDAR, it is also open to a certain rate of latency and errors.

EPC_LiDAR_car_for_print_CMYKThe primary obstacle for adoption of LiDAR is cost. Initially, units from Velodyne, the leading manufacturer of LiDAR systems, cost about $75,000 each. Where this may be acceptable for experimental vehicles, few consumers would be able to afford such an expensive upgrade. Over the last year, however, several manufacturers, including Velodyne, have announced price breakthroughs in LiDAR that promise units to be priced below $250.

Once systems reach that price level, LiDAR should see adoption on a broader range of vehicles, from semi-autonomous to full autonomy. Eventually, however, prices will need to fall even further, perhaps to $50 or less, for automotive manufacturers to consider them as sensors for all passenger cars.

The race between tech giants

Many large tech giants have jumped into developing their own driverless vehicles, but they have different goals and very different business models. For example, Waymo (Google) is developing software and systems for fully autonomous vehicles, Uber is trying to obsolete car ownership, and Tesla is trying to make electric car ownership as cool and affordable as possible.

As the technology for autonomous vehicles becomes more mature and more widely deployed, it will become less necessary to have private transportation. In that scenario, Uber may have the network advantage from its existing ride-hailing business to succeed in the new automotive economy.

When compared with established car companies such as Toyota, Ford, or GM, Uber has less to lose from the reduced demand for cars and therefore the most to gain by aggressively promoting the shared vehicle economy.

In addition to a continuous learning process, the most pressing technology that will need widespread deployment is vehicle-to-vehicle (V2V) communications

Waymo, as the pioneer in the technology, probably also has the most to gain from its deployment. While it isn’t clear exactly what business model the industry will finally choose, Waymo will be able to profit from the software derived from its deep learning, the hardware they have developed, and the mapping infrastructure that underpins navigation in general. Waymo derives no benefit from the existing automotive markets and ridesharing systems, so they have little to lose.

Tesla will make money if they can make cars that individuals will want to own and can afford to own.

What more is needed?

In addition to a continuous learning process, the most pressing technology that will need widespread deployment is vehicle-to-vehicle (V2V) communications. These systems, which have been used by aircraft for decades to avoid mid-air collisions, need to be installed on all vehicles to provide surrounding vehicles with information on speed and direction. This added bit of information can fill the holes in the current state of deep learning and make the roads a safer place for automated, as well as non-automated vehicles.

Whoever wins, it’ll be the success of the sensor and communications technology behind the systems which decides the outcome.