For years, automakers, rideshare companies, and technology firms have struggled to develop a system that allows automobiles to operate autonomously. Notably, Uber recently received a $1 billion investment to improve its self-driving tech but has yet to produce a viable solution. However, the corporation is now in talks to buy a startup called Foresight AI that might provide a technological breakthrough.
Why Uber Wants to Buy Foresight AI
Founded in late 2017, Foresight specializes in designing software that optimizes self-driving artificial intelligence (AI). However, the startup utilizes a unique methodology in pursuit of its goals. The firm deploys drones to record pedestrian and vehicle traffic at various intersections. Using that data, the startup constructs 3D digital environments akin to video game worlds in which to train autonomous vehicle AI.
By running self-driving programs through endless iterations of real-world driving conditions, Foresights uncovers an autonomous vehicle AI’s blind spots.
Despite spending $600 million a year on the project, Uber’s self-driving segment still has serious problems. Last month, the National Transportation and Safety Board reported the firm’s robotaxi software couldn’t detect jaywalkers. Accordingly, the rideshare company is wise to increase its simulation resources.
The Information states neither Uber nor Foresight has confirmed ongoing negotiations. However, the publication notes it likely won’t be as high dollar as the purchaser views the transaction as an “acqui-hire.”
It’s also worth noting that Foresight won’t be Uber’s first simulation software purchase if the deal goes through. In June, the corporation snapped up a startup called Mighty AI that makes software that optimizes computer vision programs.
A Multifaceted Approach to Autonomous Vehicle Operation
In the past, much of the discussion surrounding self-driving vehicles have revolved around road-testing software and developing higher performance chipsets. Indeed, Waymo and Lyft expressed great enthusiasm about partnering to launch a robotaxi pilot program to Phoenix, Arizona, in May. Similarly, electric carmaker Tesla boasted it manufactured vehicle operation processors capable of performing 144 trillion operations per second in April.
However, none of those companies has introduced a self-driving solution capable of fully autonomous operation.
As such, it seems there is a piece missing in the current approach to making self-driving cars. Next-generation hardware and software are part of the equation, but field-testing will provide a near-term solution. Instead, self-driving car companies might be well served by making their software as robust as possible through simulated operation.
Notably, Amazon utilized simulation software in developing its autonomous delivery vehicles. Last year, the corporation initiated a program to map an entire Seattle suburb. The firm hopes the datasets from that trial can help train its Scout drones to navigate the real world better.
If America’s largest e-commerce company believes in virtual field training for its self-driving vehicles, companies like Uber would be smart to follow its lead.