Where is my self-driving car?

Investment into self-driving vehicle technology remains high, even for pre-seed companies, as evidenced by Heex Technologies, a French deep tech startup that provides data management services for autonomous vehicle developers, raising €3.2M to support product R&D and acquire its first customers. Meanwhile, Amazon has begun testing its robot taxis on public roads in California. Despite this, driverless technology is yet to fully materialize into public life, leading many to wonder if the self-driving future will ever actually be upon us.

Undelivered promises of an upcoming transport revolution

Despite Silicon Valley’s promises, countless billions of dollars being poured into the technology, and the use of machine learning and artificial intelligence, it has been difficult to bring a product to fruition. Tesla CEO Elon Musk has been particularly vocal about the driverless future, saying just this January that he would be ‘shocked if we do not achieve full self-driving safer than a human this year’. Critics would note that in 2019 he promised Tesla would be self-driving by 2020, and he has a string of similar promises that date back to 2014. Autonomous vehicles then appear to be only a few years away, as they have been for the last decade.

Musk might be the most vocal proponent of full self-driving vehicles, but he is far from alone. Alphabet, the owner of Google, has some 2,500 employees working on Waymo, while Amazon spent $1.3 billion to acquire Zoox, and even traditional car manufacturers like General Motors have stakes in the sector, having poured resources into the acquisition of Cruise.

The problem of ‘edge cases’

Self-driving car software is a combination of ‘hard’ rules such as stopping at red lights bolstered by  ‘machine-learning’. The hard rules, making the car follow the line of the road, remaining on a certain side, and avoiding crashing, are coded in, and fairly straight-forward. The machine-learning algorithms process vast amounts of data so it can ‘learn’ to drive in the way a human can. However, while the majority of self-driving programming works well – such as a self-driving car on the highway – the rare and unusual events that occur on the road, known as ‘edge’ cases, are proving to be exceptionally complicated to code around.

Edge cases, such as a ball bouncing across a road followed by a child, a cow sitting in the road, or even complicated roadworks only rarely appear in self-driving data, meaning the car doesn’t ‘learn’ how to respond. The problem though, is that these ‘rare’ cases are not rare at all. While they happen infrequently to any one driver, when averaged across the world they happen very frequently in total. The problem is made worse because self-driving cars lack the context and situational awareness developed by humans, making them ‘confidently wrong.’ Where a human may see a ball roll into the street and slow down simply because it is an unusual situation, once a self-driving car decides something is not a hazard, it returns to normal speeds.

Better data, not more data, could be an answer

One theory on how to solve this problem is by working with different datasets. Heex Technologies believes in the need to transition from accumulating all possible data to using only the data that is needed. Essentially, that autonomous driving means companies don’t need more data to work with, but rather that they need better data. Heex Technologies has developed an event-driven approach to automate the push of relevant data (the Smart Data) to the right teams within an organization, and in order to develop better autonomous vehicles, Heex Technologies assists development teams with a more efficient and faster development process.

Setbacks as Tesla’s self-driving is recalled

According to a McKinsey & Co. report, investors have spent around $100 billion on the field of self-driving cars, so it is perhaps unsurprising that those developing products do not wish to be too morose in the future. Nonetheless, the news that Tesla will recall more than 362,000 US. vehicles to update the Full Self-Driving Beta software due to the US. regulators deciding that the driver assistance system did not adequately adhere to traffic safety laws will certainly be known and felt within the industry. It comes after Ford and Volkswagen announced last fall that they would shutter their Argo AI self-driving unit completely.

So for now, at least, self-driving vehicles will remain a few years away.

Photo : forbes.com