Technological Gaps Holding Back Autonomy
Every automotive tech show features multiple representatives talking about our imminent autonomous future. Next month… Next year… 2025… 2030… The underlying message is always the same — autonomy is on the verge of being pervasive, and your children won’t even need a driver’s license. And yet, the goal remains perpetually elusive. Sure, there are test fleets in Phoenix, Las Vegas, and numerous other major metropolitan areas. But none of them are just quite there.
Why Are Autonomous Vehicles (AVs) Not Yet an Accepted Fixture in Mobility?
There are several reasons why autonomous vehicles are not yet an accepted fixture in mobility. One of the main reasons is that this technology is still not ready. This applies both to the maturity level of the existing technology as well as to situations that autonomy may never be able to handle.
In May 2021, YouTuber JJ Ricks posted a video showing himself in a Waymo Self Driving taxi. The cab was stuck because there were traffic cones in the road. The technology produced by a multibillion dollar company was thwarted by some orange rubber on the street.
The good news is that we don’t have to worry about Skynet or Terminators because we can stop them with a few strategically placed traffic cones! The bad news, of course, is that even after years of work, the collaboration of brilliant minds, and a sum of money that beggars the imagination, traffic cones remain a challenge. And it doesn’t stop there. Issues facing autonomous vehicles include construction zones, passengers in distress, and missing street signs.
“You shall not pass.” – Traffic cones, the natural enemy of autonomous vehicles.
Eventually, these issues will be fixed and such technological shortcomings will ultimately be resolved. However, interim solutions are needed to overcome these shortcomings in the near term. If we can send people to the moon surely we’ll find a way to meet the traffic cone challenge.
Are There Situations That Autonomy Will Never Be Able to Solve?
Not everything in life is predetermined, mapped, and identified. An autonomous truck can receive digital instructions to park in slot 12 and, if there is a last-minute change, switch to slot 54. But what if the truck rolls up to a construction site? How do you direct a computer to navigate toward “the tree next to that other truck over there”? How does a tech stack respond to a rider having a heart attack? What if a passenger is alarmed by the route being used and wants an explanation?
Construction areas are not digitally mapped since they are temporary.
If Only That Was The End of The Challenges
First, there is a chicken and egg situation facing autonomous deployment. In order for AV companies to deploy en-masse and in a scalable way, they need Tier 1 manufacturers to customize and produce compute stacks and other hardware necessary for the vehicles. However, in a business where even 100,000 units are only a drop in the bucket, AV companies cannot yet generate the necessary demand to bring the Tier 1 manufacturers into the game. But if they do not have the necessary hardware pieces, they cannot deploy large fleets because the current compute solutions are not scalable enough.
Next is the network connectivity. 5G is not something anyone can rely on being available. Normally this is not a big issue for individual users. Our apps and streaming services do not overload the connection regardless of what is out there. However, autonomous vehicles need to transmit much larger amounts of data over the same public 4G LTE networks. This can be problematic and prevent critical over-the-air updates or information transmissions from happening when needed.
Finally, there are various areas where natural language processing (NLP) is supposed to fill the gap but is unable to do so. The idea is that NLP, which is a field in machine learning with the ability of a computer to understand, analyze, manipulate, and potentially generate human language, would allow for the AVs to communicate with passengers, site workers, and anyone else. However, even with mountains of data crunched through these algorithms, it cannot comprehend the necessary situations to the extent needed.
Why There Must Be a Human in The Loop
Clearly, humans must be in the loop. Humans will always be in the loop. A November 2020 report shows how almost every government regulation requires it. To have a vehicle, be it one ton or fifty tons, driving around without a human on hand to step in and assist is not only gravely irresponsible, it is impossible.
The current, and an impractical solution is to use safety drivers. These are people who sit behind the wheel of the self-driving vehicle waiting for a situation to arise for which their intervention is needed. Most of the time their job is to wait uselessly as the car drives around on its own. But what is the point of a self-driving car if you have to put a driver behind the wheel anyway? All this amounts to taking what we already have – a human driver – and planting him or her into a grossly overpriced piece of equipment.
Human in The Loop But Still Cost-Effective
Simply put, AVs need Tele assistance and teleoperation. This is when a human can remotely monitor, assist and even control an autonomous vehicle. With a near real-time video feed the teleoperators can quickly become aware of the context and surroundings which is necessary for them to implement the proper intervention needed.
This would achieve the same result as a safety driver but in an efficient and cost-effective manner. The reason why tele-assistance and teleoperation are more efficient is because most of a safety drivers’ time is wasted. They sit and wait for the vehicle to indicate that it needs assistance.
By removing these drivers from the vehicle and putting them in a control center, a single operator can be responsible for overseeing and assisting multiple vehicles simultaneously. Even if they are ONLY responsible for two vehicles, they have already doubled their productivity.
By increasing the productivity of their fleets, companies can deploy more vehicles and thus create the demand needed for Tier 1 manufacturers to get involved. By having a human in the loop there is no need to rely on NLP to solve every situation when interacting with a human. By taking advantage of the technology that keeps a human in the loop remotely, the connectivity for the vehicle can be greatly improved and all of those critical updates can be made when needed.
A fair question is whether this just creates more problems than it solves. After all, how can anyone control an AV from a remote location as effectively as someone in the vehicle itself?
The answer is compelling:
- Each teleoperated vehicle is outfitted with multiple cameras streaming smoothly over 4G LTE networks to allow for a full field of view.
- There are sensors already installed for autonomous driving purposes which can be utilized for safe remote assistance and operation.
- Remote control is ideally performed not by driving the vehicle manually but by giving high level commands. By giving a remote operator the most real time experience possible and maximizing every safety capability possible, AV performance gaps can be closed, today.
Compensate for not being in the vehicle with collision warning systems
If AV companies ever want to deploy their vehicles in a meaningful way and actually have an impact, they must implement remote assistance. Only by having a human link ready on an as-needed basis can people be kept in the loop and solve situations for AVs in a scalable way.
What Are The Limitations of Autonomous Cars?
The short-term limitations on AVs include properly identifying obstacles, understanding temporary traffic pattern changes, and inclement weather. There are long-term/permanent limitations including giving directions to unmarked locations, communicating with passengers, and other functions that require human judgment.
How Can Autonomous Vehicles Be Deployed Safely And Effectively?
In order to reliably and responsibly deploy autonomous vehicles, they must have a human kept in the loop.
What Is The Most Efficient Way to Keep a Human in The Loop With an AV?
The best way to keep a human in the loop is with remote vehicle assistance, aka teleoperation.
Amit loves marrying technology with customer needs and has been doing so over the last 14 years. Before founding Ottopia, Amit was Head of Product for Microsoft’s leading cyber-security offering, VP Product at a company building low-latency wireless video solutions, and Head of a Cyber-Security R&D department in the IDF’s 8200 Unit. Amit is also a graduate of the prestigious Talpiot program.