A rather damning Reuters report tells the tale of a long road ahead for General Motors’ and Cruise Automation’s self-driving car program. According to interviews with eight current and former GM and Cruise employees, the program has been subject to unforeseen roadblocks.
One unnamed GM source explicitly said “Nothing is on schedule,” which paints GM’s 2019 timeline to roll out a commercialized self-driving car service rather bleak. The automaker has been outspoken about launching a ride-sharing service of some sort, and top executives have reiterated the timeline. However, Cruise CEO, Kyle Vogt, told Reuters in the latest report the company is still on track to reach the 2019 goal. Those close to the project recognize it will still take a lot of time, money and resources to actually launch a fully operational ride-sharing service.
According to the numerous sources, GM Cruise self-driving cars still have major trouble identifying whether an object is in motion or stationary. One current and three former Cruise employees said they’ve witnessed the problem first-hand. For example, the car is unable to identify if bicycles parked on a rack are actually standing still. The confusion causes the car to hesitate and stop while passing parked bicycles and motorcycles.
Other software qualms, according to current and former employees, include failure to recognize pedestrians (a big safety issue) and identifying bicycles that aren’t actually present. The latter has caused GM Cruise self-driving cars to act erratically and slam on the brakes. The self-driving cars also don’t have the ability to respond to fire truck sirens.
The software robotics Cruise has used to develop the technology also reportedly slows messages from the car’s sensors to the brain, which can cause delayed actions. Vogt said next-generation software and sensors will fix that problem.
GM has promised it will only launch a service in 2019 when it meets internal safety goals. Today, those are far from reality. Both GM and Cruise have missed self-imposed deadlines and goals since the two joined forces. Missed goals include having Cruise self-driving cars record the fewest number of human takeovers from the software and it did not log one million miles by early 2018 as intended.
However, the companies said the mileage goal was based on an old footprint when more testing was underway in Arizona. Today, most of it takes place in San Francisco, where fewer miles are driven.
GM President Dan Ammann said the automaker is currently in a “race to the starting line” and conceded it will take years and billions of dollars to fully develop self-driving cars. BMW R&D chief, Klaus Fröhlich, had a different take.
“Everyone in the industry is becoming more and more nervous that they will waste billions of dollars,” he said.
Comments
Maybe they should instead focus on bringing there reliability scores up and fixing all the recent mistakes with the 2019 trucks, a controversial Camaro face and MCE’s that didn’t fix what was fundamentally wrong in the first place.
If you’ve seen one of these cars at an intersection, their software acts if it’s confused on what to do as the vehicle will sit then like a child move forward a few feet at a time trying to determine whether it is safe to do so while the drivers behind the Cruise vehicle are going nuts.
From what I read, most people giving “comments” – always anonymously, so no way to check sources – were former employees, so people with little to no clue on the current status of development. And as with all software development, there will be a backlog of features which must be fixed. That’s the purpose of testing: find stuff that does not work, and then find ways to make it work.
It it would be easy, all car manufacturers would have running prototypes today, but only a few have. The Reuters report could as well have targeted Waymo or any other self-driving development program: they will all have their fair share of issues and have a substantial backlog of things to fix. The real question is: who benefits of specifically targeting GM Cruise?
If you have looked into autonomous driving, you would know how difficult of a problem it is to solve, and how most reports are over promising when this will occur. There are first hardware issues, costs, power requirements, processing power. Then you have software, which can be accurate 99% of the time, but that 1% is incredibly difficult to create an algorithm for. So what can a car do if it’s confused? Slow down, stop. I would expect the first autonomous car services to be limited to highways (ala super cruise) or almost cars on a rail in a very narrow Geo fenced area. Expect people to screw with these things, expect the car to stop alot and slow expansion to a wider area etc. Also concerned about the political fallout when it kills someone when there is no driver present.