We may never get to Level 5 autonomous vehicles and why that is ok.
The military has been using remote pilots to fly long-range drone missions for years. Employing a very similar model, remote pilots plus long-range-vision sensors attached to the autonomous vehicle will make a vehicle’s navigating all situations autonomously (Level 5 autonomy) both unnecessary and a worse rider experience.
And it’s a good thing that Level 4 will be a sufficiently good solution, because I find it hard to imagine a world wherein the vehicle can analyze 100% of the scenarios it is “seeing” and/or “hearing” both inside and outside the vehicle.
Here’s how the remote autonomous vehicle pilot system works. Long-range vision sensors, which are looking hundreds of feet ahead, analyze the driving environment well before the vehicle is within “acting/decision range” of that physical environment. Data coming off the sensors is run through the vehicle software as normal. When/If the vehicle software is confused about how to handle the situation/environmental scenario (e.g. a policeman has set up a barricade and is directing traffic the opposite direction down a one-way street), the vehicle software alerts a remote pilot, who then has immediate access to all the data and imagery that the vehicle needs help interpreting. The pilot then decides on the appropriate course of action and instructs the car. The niftiest aspect of this entire remote pilot product and operations solution is that the rider may never even know when the vehicle struggles in this way. This is because the car never slows nor stops. It navigates the scenario seamlessly because the technology notifies the remote pilot solves the problem and instructs the car in advance of the vehicle reaching the physical scenario, and vehicle navigates the scenario as if it knew exactly how to navigate that scenario.
One downside of using remote pilots is that it in theory adds costs to the system and increases ride fares, but remotes pilots are a scalable solution over the long-term and will add considerable value due to increased safety. Over time, the technology and machine learning will have encountered so many scenarios that the number of scenarios a vehicle can’t navigate autonomously will decrease at an increasing rate, and the pilots-to-rides ratio and pilot-costs-per-ride will decrease dramatically over time.
For many years, we will rely on remote piloting to guide our robot taxis through the inevitable random scenario, and we’ll enjoy a safer and better overall robot taxi experience as a result. The alternative and far worse rider experience that will result from not employing remote pilots, is — at best — that the car will stop on the side of the road when it inevitably encounters a scenario it doesn’t know how to handle. As a rider, I much prefer a robot taxi service supported remote pilots.
These ideas are all my own and not in any way those of my previous employer. Please shoot me a note or add a comment with feedback or if you would like to discuss. Thanks for reading and sharing!