Science

Elon Musk Says Tesla Autopilot's Neural Net is "Working Well"

by Jack Crosbie
Getty Images / Pascal Le Segretain

Tesla’s Autopilot system has had a few bad trips over the past year, but as of Wednesday evening, Elon Musk says everything is pretty much hunky dory. Musk tweeted that the system’s “vision neural net” was now “working well,” but needed much more testing on the road.

Musk is big on the concept of “neural networks,” or computer systems set up to act like the human brain. He thinks that a form of “neural lace,” in which artificial systems are combined with the human brain to enhance its power, could be the way that humanity stays in control of artificial intelligence. In Tesla vehicles, the neural network just applies to a group of systems that work together in a structure that tries to mimic the flexibility of the human brain. The neural net in Tesla’s self-driving vehicles mostly regulates the cars’ guidance system — in other words, it takes a brain-like computer to make sure the cars don’t crash.

Tesla cars use a combination of cameras, radar, and other sensors to mimic, and even surpass the observational abilities of a human driver. Eventually, Tesla’s Autopilot system should be able to drive better than humans will, but right now Musk is still perfecting the combination of systems. Here’s how Tesla’s website describes the neural net:

To make sense of all of this data, a new onboard computer with over 40 times the computing power of the previous generation runs the new Tesla-developed neural net for vision, sonar and radar processing software. Together, this system provides a view of the world that a driver alone cannot access, seeing in every direction simultaneously, and on wavelengths that go far beyond the human senses.

So far, Musk says it’s going pretty well.

The “wide range of environments” part is the key here — each of Autopilot’s guidance systems has advantages and disadvantages — radar is good at picking up objects through things like clouds that would block cameras, but the cameras are best at tracking other hazards. The neural net’s success means the various systems are working well in tandem with one another, but as Musk says, it’s going to take millions more miles of stress testing before we really know how they’ll hold up in every situation.

Related Tags