Apple's Secretive Car Program Might Have A Way For Autonomous Cars To Spot 3D Objects

Render credit Jason Torchinsky. Not the real Apple Car.
Render credit Jason Torchinsky. Not the real Apple Car.

Apple spent most of 2016 coming to grips with the reality that it couldn’t build a car, but this year, the tech conglomerate showed it still has ambitions to develop autonomous driving technology. Now, a layer of the company’s famed secrecy has been peeled back, as Apple researchers have published a paper on how it believes self-driving cars can better spot cyclists and pedestrians with fewer sensors.


The paper was first noticed by Reuters, which said it was first submitted on Nov. 17 to an independent online journal called arXiv. It highlights a common feature of self-driving cars, LIDAR, a laser-based radar that essentially allows an autonomous car to see the road ahead of it. Typically (unless you’re Tesla, which doesn’t think LIDAR is necessary), this works alongside a series of cameras to capture a fuller-picture of the road, as LIDAR data is delivered in a low resolution.

In particular, the paper—which you can read here—outlines how Apple has developed a new software approache dubbed “VoxelNet.” Apple declined to comment when reached by Reuters, but the paper indicates that researchers with the company believe they have achieved “highly encouraging results” for getting self-driving cars to spot pedestrians and cyclists with LIDAR data alone.

Tests were only conducted on a computer simulation, but Apple said it was able to outperform “state-of-the-art LIDAR-based 3D detection methods by a large margin.”

“Furthermore, our network learns an effective discriminative representation of objects with various geometries, leading to encouraging results in 3D detection of pedestrians and cyclists, based on only LiDAR,” the paper stated.

Earlier this year, Apple secured a permit to test self-driving cars in California, signaling it fully-intended on keeping up in the race to develop autonomous vehicles. But this is definitely interesting. Cheaper LIDAR models are currently being developed and could come to market soon.


It almost makes me wonder if Apple’s toying with the idea of branching out and developing LIDAR on its own.

Senior Reporter, Jalopnik/Special Projects Desk


There are still challenges with this though. Telling the difference between objects that have a similar profile is the big one. I was talking to an agriculture group the other day at a conference and they had a truck that drove itself through an orchard. The main problem? It couldn’t tell the difference between a weed and a post in the orchard. You can run over a week without issue, you can’t run over the posts.

To put this in self driving car perspectives, say you have a tumbleweed or a piece of litter blow into the road. What is an autonomous car going to do? If it sees a huge sheet of newspaper is it going to know it can drive through it or will it think it’s a giant solid object that it needs to stop for? If some jerk on the sidewalk throws a cardboard box in the road, will it stop all traffic until someone walks out and moves it?