AI researcher and podcast host Lex Fridman, has interviewed some of the top minds in the field of autonomous vehicles and AI.
Lex teaches Deep learning at MIT and also focuses on human to robot interaction. While his interviews are fascinating and entertaining, his latest video is from a drive he took with Comma.ai founder George Hotz.
Throughout the 17 minute video, Lex questions George on Comma’s functionality, what works well, what doesn’t, which provides insight into where the platform is currently at.
Comma.ai uses computer vision to understand the environment around it, however, the solution is sold as an aftermarket add-on to existing vehicles. Commai.ai’s approach to autonomous driving uses driver monitoring to track the alertness of the driver, which enables hands-free driving.
As an increasing number of autonomous driving solutions move to offer hands-free driving, by implementing driver monitoring, it does put pressure on Tesla to offer the same, particularly if they want FSD and Autopilot to be considered the best in everyway.
Currently, Tesla’s requires the driver to feed pressure into the wheel registering the driver is there, paying attention through a torque sensor. Reminders to provide this data point to the car, occur every 30-40 seconds.
Currently GM’s SuperCruise offer’s hands-free driving, as will Ford’s upcoming tech in the Mach-E with Active Drive Assist. With Comma.ai also in the race, I feel we’re pretty close to Tesla needing to respond to this. While Elon may be keen to leapfrog the need to pay attention at all, it seems promises we’d see Level 5 by the end of 2020 are far from reality.
One example is when Hotz details his favourite feature – long trips. He explains that you just sit there and watch, with the technology currently at a point where a drive from San Diego to LA (around a 2-hour drive) can occur almost without intervention.
Some other attributes of the Comma.ai solution is that it doesn’t require lane markings to work. In fact, it can even perform lane changes, on sections of roads without lane lines.
We do see Hotz needing to disengage to take a highway exit ramp, so that’s certainly one area where Tesla seems far more competent. Comma.ai also is yet to understand traffic lights and stop signs, while Tesla’s have been capable of that for months now.
After the drive, Fridman takes a tour of the Comma headquarters in San Diego. During the tour, the pair review analytics on large displays in the office. An important metric is that Open Pilot has now accumulated more than 30 million miles of data. This data feeds the AI models that enable the car to navigate. Essentially the more data you have, the better the autonomous driving solution becomes.
As Hotz showcases the user data, the camera pans to a display showing reports on usage in the past 30 days. Naturally the US ranks highest, followed by Korea, Canada, Taiwan, China and then Australia, with 42 users in the past month, having driven 3,893 routes during that time.
Diving deeper into the data, it seems those who have Comma.ai installed, they engage it for half of the miles (or km) they drive, or in time portions, around 30% of the time. This suggests users are find pretty good benefit from what’s on offer today (lane centering etc), but there’s some way to go before it reaches a level 3 system.
Now for the video, watch and enjoy.
Consumer Reports recently released a report into Active Driving Assistance Systems and ranked Comma.ai first, ahead of GM SuperCruise, followed by Tesla Autopilot.