Can Tesla and other fully autonomous cars really work in the real world?

Man working on laptop & texting on mobile in self-driving Tesla Model S car on autopilot (Source: ShutterStock)

The new Mercedes-Benz S-Class is a hoot. It features one of the highest levels of automation found on a production Merc yet, and that includes its ability to drop you off at the curb and go looking for a parking spot, all on its own. While the feature, along with several others, hasn’t been homologated for the Indian car market, it does bring into sharp relief the role played by autonomous software in driving the technology of the future.

While the S-Class has been, for the longest time, tasked with bringing the future to present-day cars, in this regard it seems to have been beaten by the likes of Tesla, whose own Autopilot system is capable of helping you navigate off highways, respond to traffic lights, stop signs and be summoned from a parking lot, through a mobile app.

Seems like the future has arrived, except, for some of Tesla’s customers, the brand hasn’t fulfilled its promise of fully autonomous driving. With mounting lawsuits over their claim of full, functional autonomy, Tesla has been forced to reconsider just what sort of real-world self-driving capabilities they can and will continue to offer. At present Tesla charges its customers a large sum, going up to Rs 7.2 lakh for what is mistitled as its “Full Self-Driving” or F.S.D package.

In light of the constant scrutiny that Tesla’s Autopilot seems to be under, particularly from government bodies, CEO Elon Musk appears to be singing a slightly different tune with regards to its feasibility than he was known to in the past. On June 6, Musk tweeted “Generalised self-driving is a hard problem, as it requires solving a large part of real-world AI. Didn’t expect it to be so hard, but the difficulty is obvious in retrospect”.

Having realised the hard way that fully autonomous driving capabilities are extremely difficult to achieve, if not a pipe dream altogether, Musk is practising caution with his claims of full self-driving capabilities, admitting that “nothing has more degrees of freedom than reality”.

So what exactly are the factors that are holding back fully self-driven cars? And should human intervention remain a necessary part and legally mandated aspect of operating what is essentially a metal projectile?

Where are we with autonomous tech at the moment?

It isn’t surprising that the key developments in the race towards fully autonomous cars are coming out of Silicon Valley with Tesla, Alphabet and Apple in the lead. At present, self-driving tech primarily relies on lidar and radar technology to track the speed and distance of nearby objects. Tesla, however, prefers a different approach, having phased out radar tech from its cars, relying instead on cameras to do the job.

It doesn’t appear to be working all that well. The most recent of Musk’s confessions came on August 24, when he confessed that the latest prototype for Tesla’s patented FSD Beta 9.2 “isn’t that great”. Safety experts from Consumer Reports, an American not-for-profit organisation that independently tests consumer products, expressed concern regarding the prototype after Musk tweeted “There will be unknown issues. So please be paranoid”.

Also Read: Musk says Tesla’s autopilot software is ‘not great’

Once again, FSD Beta 9.2 is something of a misnomer as it doesn’t promise full self-driving capabilities. In an age of crowded highways and gridlocked streets, the allure of self-driving cars can be strong, given that they come with the promise of obliterating driver stress and not just significantly reducing it.

But driving does not only consist of switching lanes on the highway or parallel parking. It requires constant spatial awareness, risk assessment and most importantly constant reasoning which a car cannot or more importantly should not be capable of doing. However, the debate over self-driving seems to have taken a few steps back from liability to actual feasibility.

At present, car brands are unlikely to remove the need for human intervention from their self-driving modes, seemingly to avoid further legal hassles. However, it’s also a question of capability. According to an article in The New York Times, camera technology and image processing – something that Tesla relies on completely for its Autopilot and FSD feature  – isn’t quick enough to avoid crashes. The article further states that the technology to do this safely and consistently does not yet exist.

Also Read: 'It happened so fast': Inside a fatal Tesla Autopilot accident

Even Apple appears to have hit a speed bump with their autonomous car project, with multiple project managers having quit in recent times. According to a Bloomberg report, Apple’s self-driving car project has had to hit the reset button multiple times since it was announced back in 2014. And while Apple is allegedly busy hiring top talent from the car industry, the project isn’t likely to hit the road at least until the end of the decade and not 2024, as it was earlier intended to.

Legislation yet to catch up

While brands like Audi and Mercedes-Benz also have advanced AI capabilities installed in their top luxury models, getting the legislative greenlight has proven to be a challenge. Back in 2018, Audi had already equipped its flagship luxury sedan, the A8 with Level 3 Autonomous tech. At Level 3, Audi’s accurately named Traffic Jam Pilot was equipped to take control at low-speeds and heavy traffic but was happy to hand over controls once things got more interesting. While the German government amended its highway laws to accommodate Level 3 technology, many other countries, India and the UK included, did not.

According to the UK’s Road Vehicle Regulations, no driver was allowed to relinquish driving duties to someone who could not have a full view of the road ahead. The law did not take into account the possibility of that someone being something. This proved to highlight once again, that the technology for autonomous tech was developing much faster than the legal framework required to keep it in check.

At present, laws are designed to assume that a human is behind the wheel, which is the reason brands like Tesla, Audi, BMW etc are ensuring that human intervention is conditional to the operation of AI tech in their cars. It’s not just the laws but the roads and driving infrastructure that’s been designed for human vision. For fully autonomous Level 5 AI to work would require a total infrastructural overhaul.

At present, the autonomous car market is valued at $54 billion. And while a great level of interest is coming from fleet operators and small businesses, it hasn’t stopped legacy carmakers from investing heavily into the technology. From the looks of it however, it appears that autonomous tech might be limited to mass modes of transport, with fixed pathways.

Much like Waymo’s self-driving robotaxis which have begun being field tested in San Francisco. For something as complicated as a privately-owned passenger car, with a multitude of constantly changing variables strewn in its path, the road to full autonomy is yet to be paved.