Britain begins to embrace Self-Driving technology
The dream of jumping into a car without a driver behind the wheel and being chauffeured to your destination is incredibly enticing, not only in terms of comfort and productivity, but also safety. As the technology is not susceptible to distraction from mobile phones or drowsiness and ideally has perpetual 360° vision of the surrounding area, allowing it to constantly monitor nearby traffic and spot possible dangers much more proficiently than any human ever could, as despite how careful a driver you are the real threat comes from other road users and their unpredictability.
The standard model of autonomy has a graded scale with levels of increasing functionality, with level 0 being a traditional vehicle requiring a human to perform all the work and level 5 being fully independent of human interaction, able to handle every situation it encounters and allowing the occupants to completely disengage from the often monotonous and stressful task, providing the perfect opportunity to catch up on emails or remote work, pop on a movie or your favourite TV show, or dive into a video game.
Industry standard levels of autonomy - Source: Society of Automotive Engineers (SAE)
The UK government recognises that once the technology is proven to be successful and safer than the average human driver, this is the future that everybody will want access to. In preparation for this, they've recently released proposed changes to their motoring regulations - the Highway Code. It raises questions such as how to handle ethical dilemmas (see the infamous trolley problem) and who will be responsible in the event of an accident and these will certainly need to be thoroughly discussed.
'...the government expects to have a full regulatory framework in place to support the widespread deployment of the technology by 2025, helping to make the movement of people and goods safer, greener and more efficient. The technology could improve road safety across Britain by reducing human error, which is a contributory factor in 88% of all recorded road collisions.'
Many companies believe they have what it takes to reach level 5 albeit with differing approaches, the use of LIDAR is commonplace and whilst providing benefits in specific situations, it may not be required or even useful for these systems to operate with a success rate higher than any human could achieve.
Self-driving vehicles using this technology have been around for many years with projects such as Google's Waymo and San Francisco based Cruise, but currently they're locked within strict, pre-mapped boundaries and frequently display bizarre behaviour. This is inherent to that method of achieving autonomy and known within the development community as a 'local maximum', where the performance gains have peaked and diminishing returns ensue. Whilst the progress so far is impressive, there's little room to improve from here and you cannot simply drop the car on any road in the world and expect it to work, which is clearly the ultimate aspiration, so it seems a different approach is necessary.
As Tesla rightfully states, the road network has been designed for processing by an organic neural network and a pair of optical sensors (i.e. two eyes and a brain) and thus all that should be required is stereo cameras and an extremely competent and well-trained neural network software. They've been working on this problem for years now through multiple generations of hardware; from third parties MobilEye and NVIDIA, finally deciding that nobody provides an off-the-shelf solution capable of meeting their needs and building not only their own processing chip from the ground up, but also the most efficiently scalable, modular supercomputer system named Dojo that once completed will be used to automatically train the software by AI machine learning. Thanks to input from hundreds of thousands of users of their BETA program who are as we speak playing chaperone to their vehicles as they slowly learn the many complexities and operations required to navigate our often confusing and imperfect traffic systems, sending countless terabytes of data to be analysed for future updates, an obvious advantage of massive scale fleet-learning that so far only they have been able to deploy.
With tens of millions of miles driven and an average safety rating already over 4 times higher with Autopilot than organic life, the potential for this life-saving technology is self evident and clearly illustrated in their public safety data comparing it to the average US driver.
Source: Tesla’s quarterly safety report - (note: as of Q1 2021 they stopped sharing statistics differentiating driving without AP enabled)
Not to say that multi-billion dollar companies are the only ones having success, with entrepreneurs such as George Hotz from Comma.ai currently producing a kit you may install yourself, starting at just $2,000 it’s compatible with 150+ vehicles and provides basic levels of autonomy with the same idea of just a few cameras and a computer running their Openpilot software, more are sure to enter the emerging market.
This is a big step forwards for the United Kingdom and whilst not the first by any means, it's great to see they're clearly keen to have this technology transform the economy, reduce pollution and most importantly save countless lives! I can only hope that it arrives even quicker than planned since Elon Musk expects Tesla's Full-Self-Driving to be reliably safer and more capable than a human by the end of 2022 - bold words perhaps, but coming from the only person on Earth to have developed fully autonomous, self-landing rockets, the destination seems inevitable with only the timeline in question.
The official announcement is available here: https://www.gov.uk/government/news/britain-moves-closer-to-a-self-driving-revolution.