Tesla’s fleet of electric cars will be able to recognize and react to first responders in the near future. In a recent announcement on Twitter, CEO Elon Musk noted that the company is planning on adding vehicles like police cars, fire trucks, and ambulances to Tesla’s Neural Network in the “coming months.” Such an update would further raise the safety level of Tesla’s vehicles, particularly when they are operating under Autopilot.
The behavior of Tesla’s vehicles became a point of interest for mainstream media recently, after a Model S owned by Los Altos Planning Commission chair Alexander Samek got involved in a rather unusual police chase. During the incident, which happened late November, police discovered Samek asleep behind the wheel of his Tesla, which was presumably operating on Autopilot. Police were eventually able to stop the Model S by pulling in front of the electric car and gradually slowing down, though it took seven miles before the vehicle completely pulled over. Samek was later arrested for driving under the influence of alcohol.
We’re adding police car, fire truck & ambulance to the Tesla neural net in coming months
— Elon Musk (@elonmusk) December 3, 2018
In a tweet about the incident, Elon Musk noted that Tesla was looking into what happened to Samek’s vehicle, particularly since Autopilot is designed to slow down to a stop and turn on the hazard lights when driver input is not detected. Thus, there must have been specific factors involved that allowed the Model S to continue driving for miles before it pulled over. A likely scenario would be Samek inadvertently putting pressure on the steering wheel while he slept.
Neural Net improvements would likely prevent similar incidents from happening. Tesla’s Neural Network, after all, continues to evolve over time and as more vehicles are deployed on the roads. Over the years, the electric car maker has trained its Neural Net to recognize specific vehicles, such as motorcycles and trucks. If Elon Musk’s tweet is any indication, it appears that Tesla has also all but trained its Neural Network to recognize first responders as well.
Exactly. Default Autopilot behavior, if there’s no driver input, is to slow gradually to a stop & turn on hazard lights. Tesla service then contacts the owner. Looking into what happened here.
— Elon Musk (@elonmusk) December 3, 2018
Elon Musk’s rough timeline in his recent tweet seems to be well in line with Tesla’s expected release for Hardware 3, which would feature the company’s custom-designed Autopilot computer. Elon Musk has been optimistic about Hardware 3, dubbing it as the “world’s most advanced computer designed specifically for autonomous operation.” Designed by a team led by Pete Bannon, who helped create Apple’s first ARM 32-bit and ARM 64-bit processor, Hardware 3 is expected to be capable of computing “2,000 frames a second and with full redundancy and fail-over.” Last October, Musk noted in a series of tweets that owners who purchased Tesla’s Full Self-Driving suite would get the Hardware 3 upgrade for free.
The release of Hardware 3 would likely usher in the rollout of larger, more advanced Neural Networks. This was teased by the company during its third-quarter earnings call, when Tesla Director of AI Andrej Karpathy noted that the company has already trained large Neural Networks that function very well. Elaborating further, Karpathy noted that Tesla is unable to roll them out due to limitations in the company’s current hardware.
“The team is incredibly excited about the upcoming upgrade for the Autopilot computer. This upgrade allows us to not just run the current Neural Networks faster. But more importantly, it will allow us to deploy much larger, computationally more expensive networks to the fleet. The reason this is important is that it is a common finding in the industry, and that we see this as well, is that as you make networks bigger by adding more neurons, the accuracy of all their predictions increases with the added capacity.
“So in other words, we are currently at a place where we’ve trained large Neural Networks that work very well but we are not able to deploy them to the fleet due to computational constraints. So all of this will change with the next iteration of the hardware, and it’s a massive step improvement in the compute capability. And the team is incredibly excited to get these networks out there.”
Once Hardware 3 is rolled out and the larger, trained Neural Networks are deployed to the company’s electric car fleet, incidents such as Samek’s would likely be handled by the company’s vehicles in an even safer manner. While Autopilot most certainly saved the Los Altos Planning Commission chair’s life, a larger Neural Network that recognizes and responds to first responders would have allowed for a much faster resolution to the incident.