Autonomous-Remote-Controlled: Robotaxis Could Be Here Faster Than Expected

There is a lot of confusion when discussing the development status of self-driving cars. While German experts and car manufacturers are of the opinion that autonomous cars will not be around for 10 or 20 years at the earliest, companies in Silicon Valley and China are much more optimistic. Extensive tests with even completely driverless vehicles are already on the roads, albeit in regions with comparatively simpler traffic conditions, such as Chandler in Arizona. And then there is Teslas Elon Musk, who announced at a conference in China this week that autopilots with the autopilot are already “very close” to level 5 driving and could be uploaded to one million Teslas before the end of the year.

While progress is undeniably being made, it is happening more slowly than the public would wish – partly through promises from industry and partly through wishful thinking. In the hype cycle, development is going through the valley of tears.

As it turns out, it is quite easy to program a car to drive between lane markings. But autonomous cars have to solve a number of other problems. The detection of objects, planning the optimal route and driving itself. But once we get to the intentions of other road users and how they react to them, and take passenger comfort into account, things get hairy. The further down the list, the more difficult the problem becomes, the more we need to test and try out, and the less often these situations occur. The effort and time required increases exponentially.

Degree of Autonomy

I would like to define here the degree of autonomy of the vehicle as a factor of time and the number of average driving situations that such a vehicle can handle correctly in daily use in averagely complex traffic. I define all incorrectly detected situations as those where the vehicle comes to a safe standstill and waits for further instructions, or is manually controlled. For example, the instruction may come from the control center or the driving situation may have resolved and the vehicle recognizes a known situation to which it can react independently. This means that not every non-autonomy time requires human intervention.

With a degree of autonomy of 90 percent, such a car would come to a standstill for a total of 72 minutes or would have to be operated manually, assuming an operating time of 12 hours per day. The number of daily driving situations (which itself would again require its own definition) can be up to several thousand per day. This means that we would have at least one hundred unknown situations every day. However, a certain number of these could be solved by the car itself as soon as the existing unknown situation has changed into a known traffic situation.

Degree of AutonomyNon-Autonomy-Duration [hh:mm:ss]
Non-Autonomy-Duration of an autonomous vehicle assuming shifts of 12 hours per day

The individual non-autonomy-duration naturally depends on the complexity of the respective unknown driving situation, the numbers quoted here only serve as an average.

Expenditure Factor

I define the expenditure factor as the amount of time and money that an organization needs in order to go from, for example, 00 percent autonomy to 99 percent. Assuming that this is an exponential problem, this means that to get from 0 percent to 90 percent, for example, a monetary investment of 100 million dollars and a million kilometers driven (with the corresponding time expenditure) could be for a specific team. These figures are only examples to illustrate the problem and the definition. In order to reach a 99 percent degree of autonomy, one has to calculate with ten times the effort and expenditure. So in our example we would have to calculate one billion dollars and 10 million driven kilometers.

Degree of AutonomyExpenditure FactorTime Expenditure
[Mio Kilometers]
Monetary Expenditure
[Mio US$]
Expenditure factor to reach the next degree of autonomy measured in time (in real kilometers, not simulated) and money, calculated for a team.

Where exactly these figures lie and how they fit together depends on each individual team and on the overall effort of the industry. And this is also where the regulator comes into play, but more on that in a moment.

Let’s look at the time spent first. Currently, only one company – Waymo – has driven more than 10 million kilometers but well under 100 million autonomous kilometers in reality. Several billion more simulated kilometers are being added. All the others are below that, in some cases even far below that with not even one million kilometers.

So Waymo is probably at a – wherever we set the limits of the degree of autonomy – degree of autonomy between 90 and 99 percent. With the exception of GM Cruise and Waymo, all other of the 66 companies testing in California are below our 90 percent mark.

If there were only one team working on the development of autonomous cars – and by team I mean a start-up, a company, a research group – then that team would have to take on the entire burden and make up for both the time and material expenditure. Currently, the companies only collaborate with each other to a limited extent, although there are overlaps in the use of open source data.


A regulator could accelerate development and share the burden (at least partially) by encouraging and enforcing the sharing of critical data and algorithms across industry. Between 2013 and 2017, conservative estimates suggest that at least $80 billion has already been spent industry-wide on developing autonomous cars, but we are not even close to 99.99 percent autonomy, because many efforts were doubled by everyone else.

According to the current opinion, autonomous cars can only be used for a regular commercial robotic taxi service without a driver from a certain degree of autonomy. This is perhaps the 99.9 or 99.99 percent. Since further development – as we have seen – will swallow up a lot of money, but at the same time will not bring in any immediate revenues, because so far these are exclusively experimental vehicles, the companies are in a quandary. On the one hand, they want to bring a safe technology to the market, but on the other hand they are under pressure to finally make money with it. And this is where regulators could approve a hybrid technology approach.

Hybrid Remote-Controlled-Autonomy

The idea of using teleoperations to control a delivery robot, an e-scooter or even a semi-autonomous car is not new. Nuro, Einride, Nissan and others have already introduced these concepts. entire business model is based on a service to teleoperate of vehicles.

Now Voyage CEO Oliver Cameron has introduced the new Voyage Telessist in a blog post. This is a self-developed pod, with which an employee can remotely take control of the Voyage Robotaxi in such discussed situations. And this was not limited to robot taxis, but could be extended to trucks and delivery robots. In the case of trucks, the idea was anyway, that only the drive on highways itself was to be done autonomously, and that the drives from the distribution center to the highway and from the highway to the final destination was to be done by a driver who boards the vehicle at the ramp. This could now be replaced by this hybrid model.

Voyage Telessist

And here’s where everything falls into place. A hybrid of semi remote-controlled, semi-autonomous and driverless vehicle would solve several problems at once and offer a lot of advantages

Firstly, start-ups and companies would no longer have to wait until their cars have reached 99.9 or 99.99 percent autonomy in order to be approved as robotic taxis. They could do it immediately and would only have to prove that in non-autonomy times the vehicle would safely come to a standstill or could be taken over.

Secondly, this would also bring immediate revenues and thus cash flow into the coffers of the developing companies.

Thirdly, it would allow companies to operate a fleet of robotic taxis, while at the same time driving development forward. The more vehicles are on the roads and the more they drive, the greater the number of rare traffic scenarios that fleets encounter. The fleets can thus be improved faster.

Fourthly, because no more drivers are needed in the vehicle itself, a smaller number of drivers in the control center can remotely control the vehicles. One remote control driver is responsible for monitoring and controlling several vehicles. Initially, one driver could be responsible for 10 vehicles, but with continuous improvement this rate could then drop to, say, 1:1,000.

Fifthly, this will lead to massive cost savings, which could also have an impact on car hire costs and increase acceptance by passengers, and offer cities and regions low-cost alternative mobility services.

Sixthly, autonomous cars could thus be used as a mobile solution for the masses much earlier than expected. And these cars could also be more sustainable, since they cannot drive with a “lead foot” and in a more optimized way.

Such an approach is not an admission that pure autonomous driving is not possible. As with other technologies before, these are used before they are technically 100 percent perfect, but are already capable of solving other problems. If a safe solution can be offered that is both more economical and more sustainable, then we should do everything we can to bring it to market.

In this respect the significance of the Voyage Telessist cannot be appreciated enough. Exactly the combination of 90 percent autonomy and remote control can give a boost to the use and development of autonomous vehicle technology and make the bold assumption of the title of this blog and my book a reality: The Last Driver’s License Holder Has Already Been Born.

This article has also been published in German.


Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s