Thank you Ry, and apologies for my missing this response!
Some material to get you going on this topic: I’m sure you’ve gathered that as with any cutting-edge technology, the numbers around processing, transmission, and storage are a moving target. The data-driven world of one autonomous car in testing is vastly different than that of a swarm of consumer-grade autonomous cars interacting while also appeasing the customers and corporations involved in the operation of the fleet. Some of these areas have made highly-documented progress, and some are lagging behind, but if you’re curious, the below reads are a good start.
Acquisition: LIDAR is the digital eyeball of choice for most AV developers, and also the poster boy for Moore’s Law in this industry. When I started this blog two years ago, LIDAR was an $80,000 piece of hardware, leading every naysayer to laugh and proclaim self-driving cars impractical for consumer price points. A year later, someone made a better LIDAR scanner for $8k. About six months after that, researchers improved on the tech at a projected unit cost of ~$500, and the news from a month ago is that MIT engineers have gotten rid of all moving parts, put the remaining bits on a chip, and gotten it to a mass production price point of $10. Crazy stuff.
Processing: Folks like NVIDIA are killing it right now, building better, cheaper chips that increasingly focus on the kinds of processes required by self-driving cars. Their DRIVE PX2 chip was already arguably the gold standard for tech-heavy vehicles, and now they’re saying they’ve upped their game even further. Full disclosure, I own (a pittance of) stock in NVIDIA.
Storage: last I heard, Google’s autonomous cars are pulling down a gig every second, which of course translates to a preposterously large amount of storage required. But then we’ve got startups like Civil Maps, who claimed they were able to squeeze a terabyte of map data into an 8mb file.
Transmission: this is probably the slowest-moving piece of the puzzle. A few decades ago, the FCC — in a startlingly visionary move — declared they would earmark a chunk of wifi’s (currently unused) spectrum for intelligent transportation & infrastructure. A lot of folks worry that won’t be enough, and on top of that, you’ve got corporate interests arguing we should free up that bandwidth now for more traditional use. It’s clearly the most bureaucratic of the elements, but who knows… maybe that will inspire startups to solve the problem in other ways.
