The Self-Drive Revolution

The US players taking the wheel in the Autonomous Vehicle Revolution

Written by: Phoebe Harper


The evolution of self-drive vehicles has begun to read as a tale of false starts and delayed developments. Automotive enthusiasts worldwide wait on tenterhooks for the latest announcement from automakers at the forefront of the industry, poised for what promises to be a ground-breaking development.

For many experts within the transport sector, the invention of self-driving passenger vehicles resembles a natural stage of technological evolution, akin to the development of the mobile phone from landline to portable. Yet, self-driving vehicles are a technological advancement that has pre-occupied leading engineers and swallowed millions of dollars in investments for years.

The concept of a self-drive car has grown in tandem with significant developments in Artificial Intelligence (AI) over the course of the last decade. However, what is referred to as ‘Level Five’- the Holy Grail of a fully self-driving ‘robocar’- is yet to be fully realised, as human supervision remains a full-time requirement outside of trial settings.

2020 was originally tipped as the year that would herald the arrival of the self-driving car, as suggested by the likes of Toyota, Honda, and Waymo. For Tesla, 2020 became the fallback launch option after the company’s original forecast of achieving self-driving cars by 2018 proved overly ambitious.

Leading automotive industries have since taken a step back, with former front-runner Uber announcing that it has effectively given up on its plans of self-driving taxis, selling off its autonomous division to San Francisco-based start-up Aurora in a $4 billion deal. 

Inevitably, the worldwide grip of the pandemic has frustrated progress somewhat, as the focus shifts from prioritising robotic passenger vehicle development to autonomous delivery and trucking services to minimize risk of infection whilst providing essential logistic services.

Rumours continue to amplify surrounding Apple’s ‘Project Titan’. Initially launched in 2014, the project entails the launch of a fully automated vehicle by 2024 complete with breakthrough battery technology, as reported by Reuters. It is still unclear who the tech-giant will partner with, as rumours of a Hyundai-Kia Motors partnership were recently dispelled, leading to a plummet in shares for both motor companies. Nissan have also denied involvement in Apple’s project for an autonomous car, supposedly over branding disagreements.  

Within the restricted areas of trial zones, Tesla are one step ahead on the road towards full autonomy, after releasing the first version of its Full Self-Driving (FSD) beta software to a small group of testers in October 2020. Trials are ongoing across California to reach what Elon Musk terms, “superhuman self driving.”

While most people tend to associate self-driving cars with a fully automated experience - the driver reclining nonchalantly in the passenger seat, perhaps responding to emails whilst nursing a coffee on the morning commute – it is worth remembering that there are varying degrees of automation which offer a clearer model to chart the progress that has been made within the industry.

In terms of semi-automated offerings that are highway-ready, there are several options available. The Cadillac Super Cruise, along with the Nissan ProPilot Assist and Tesla’s semi-autonomous Autopilot system are the closest thing to the automated experience on the market. Benefiting from automated technologies, such as lane centering, semi-autonomous navigation and traffic-aware cruise control, designed to keep the cars in lane and at a safe distance from other vehicles, your hands are free to leave the wheel. ‘Level Two’ autonomy is well within reach.
_________

The Six Levels of Autonomy Explained
 
Level 0 – Zero Automation features.
Level 1 – Low level driver assistance features, e.g. lane departure assist, adaptive cruise control.
Level 2 – Advanced driver assistance systems (ADAS), partial automation but requires drivers full attention.
Level 3 – Autonomous Vehicles that can interpret the world around them without human intervention – driver should be ready to take over if necessary.
Level 4 – Technically no driver needed, but geographical restrictions apply for where this tech can be used. Top-speeds are limited, so must be driven by humans for faster speeds, e.g. Google Waymo.
Level 5 – The end-game of self driving technology – the car drives consistently regardless of road type or external weather.
_________

The simplicity of the concept of an autonomous vehicle (AV), belies the complexity behind the technology. Many would argue that the engineering has been underestimated. The advanced control systems within the vehicle that interpret essential sensory information, such as obstacles and signage in order to navigate correctly, must be developed enough to allow total reliability.

Naysayers worldwide are wary of safety concerns, and perhaps justifiably so. A myriad of variable factors complicates technological advancement and present serious obstacles to AVs – whether it’s the weather, errant jaywalkers, or the challenge of compensating for the physical communication that often takes place between drivers.

AV models rely on cameras to track external objects, allowing the car to react to avoid a collision. Alongside this, in-car computers ‘machine-learn’ the rules of the road and identify the appropriate navigation. This takes time and can only be furthered through real-world examples, meaning the cars themselves physically need to be travelling on the road. At Waymo, it took a whole decade just to clock its first ten million miles which marked exciting progress.

The number of miles driven acts as proxy for how much training data the company has collected, ultimately translating to safer vehicles.

Automakers are partnering with titans of technology to benefit from software improvements that will aid them on their journey to developing reliable automated driving systems (ADS).

Cruise, a leading self-driving car company and partner of General Motors, has joined forces with Microsoft Corp to benefit from Azure, Microsoft Cloud’s computing platform to accelerate the development and commercialization of self-drive vehicles.

In Silicon Valley, self-driving car start-up Aurora are one to watch technologically speaking. Founded in 2017 by a formidable triumvirate in the field of automated vehicles, Waymo’s founders include the former director of Tesla Autopilot, a previous Waymo engineer and Uber’s former autonomy and perception lead. Dedicated to the software, data and hardware behind the driverless cars rather than the vehicles themselves, Aurora has established itself as a leading name within the field.
_________

The Basic Technology
Radars – Essential for identifying other vehicles by emitting radio waves that bounce back from their surroundings.
Cameras – As machine vision improves, the cameras will be used to identify and navigate everything in view. At present, they are used to identify key markers on the road, e.g. traffic lights.
Lidars – ‘Light detection and ranging’ – data collected from millions of lasers used to build a 3D map. These are the spinning thing installed on top of most AVs.
Maps – All ‘robocars’ will have their territory mapped by lidars and cameras extensively by the parent company. Essential for the car to know its precise location and to use as reference to verify sensor data.
Machine Learning – The AI tool that enables computers to detect external objects, such as cyclists or lane lines, through millions of real-life examples. Through this, the cars will learn to react independently.
_________

Different companies incorporate the technology using alternate methods. At Tesla HQ, ADS are being developed through coarse-grained two-dimensional maps and cameras, combined with radar and ultrasonic sensors. Comparatively, Google's Waymo are also utilizing radar and ultrasonic sensors but instead they are relying on extremely detailed centimeter-scale three-dimensional maps and lidar. Tesla argues that although its process is more difficult technologically, it will have more benefits in the long term.

Trials are ongoing, and the software is ever developing to improve how the vehicles interpret the sensor data. The improvements that still need to be made in this area can be tracked through disengagement ratings, such as those reported by the State of California’s DMV for 2020.

‘Disengagement’ occurs when an AV is forced to disengage from autonomous mode during testing – this could be the result of a technological failure or a situation that means the test driver or operator has had to take manual control of the vehicle for safety reasons. Statistics are rated by the number of instances per mile driven. While most companies typically don’t report these numbers, in the state of California it is a requirement.

This objective report speaks volumes about the progress of different companies. Both Waymo and Cruise are statistically ahead of other prestigious competitors, including Apple. Both have marked significant improvements from their 2019 figures as flaws continue to be minimized across both companies’ autonomous driving technology. Google-owned Waymo is the front-runner, with a record of 21 disengagements recorded over 630,000 miles of testing, translating to 29,425 miles covered per disengagement.

Shockwaves were sent through the transport sector earlier this month, as on February 10th 2021, Germany became the first country to introduce legislation enabling the use of autonomous cars and buses in defined zones nationwide that are expected to be in regular operation by 2022.

The regulatory approval needed for Tesla and other autonomous carmakers to bring self-driving vehicles to public roads is one step closer – although drivers may not be able to relinquish the wheel just yet, progression is well under way.