Electric car pioneer Tesla revealed a full hand on self-driving technology at its “Autonomy Investor Day” which it held Tuesday morning (Australian time) from its headquarters in Palo Alto, California.
It is the first time that the EV maker, which began installing its proprietary self-driving AI chips in all new Model S and Model X vehicles in March and this month into all new Model 3s, has given its investors and the media the chance to deep dive into its technology.
While the original reason behind investing in the development of full self-driving technology by Tesla has always been to reduce cognitive demand for drivers, the EV maker has now also confirmed it drives a second goal – a fleet of self-driving electric “robo-taxis” that it wants to have operational by 2020.
Its plan for fully autonomous driving is by not new, it is the first time CEO and founder Elon Musk has gone into such depth into the details of the AI tech that Tesla is developing, and it has both the media and autonomous vehicle (AV) world talking.
At last year’s second quarter earnings call, Musk confirmed that Tesla had a cat in the bag; a neural network that it had been developing under the guidance of former Apple hardware architect Peter Bannon.
Put simply, Tesla had decided that instead it could improve by an order of magnitude the workings of the “traditional” neural net processing units – a combination of graphical (GPU) and computer processing units (CPU) – by starting from scratch to combine both processes in one chip – making one AI chip to rule them all, so to speak.
At the time, Musk said the Tesla AI chip would be able to process 100 times more frames than handled by the Nvidia GPU currently in use by Tesla cars.
Today’s presentation started with Bannon giving a tour of the AI hardware and who was followed by Andrej Karpathy, Tesla’s director of AI, presenting a rundown on the tightly integrated software that runs the chip.
From numbers of pixels processed per second, the breakdown of the AI chip’s components, to explanations about the security and redundancy, here are the most salient takeaways from the presentation:
The AI chip only uses 100 watts of power
While this is in part to ensure retrofitting into existing vehicles remains a low cost, keeping a cap on power usage of the chip is important for another reason: range.
The target power usage for the Model 3 is 250 watts; an increase in power would substantially reduce the range of the post-leasing Standard Range Model 3s Tesla has pegged for its robo-taxi fleet, particularly in cities where the fleets would operate.
It uses cameras, ultrasonic sensors, maps, and more – but not LiDAR
While many other self-driving tech developers are putting a huge amount of investment into using LiDAR technology, Musk says the light-sensing version of radar technology is a “fool’s errand”.
It’s expensive and unnecessary, and anyone relying on expensive sensors will be doomed, Musk said. “It’s like having a whole bunch of expensive appendices, one appendix is bad, well why have a whole bunch of them, that’s ridiculous.”
The AI chip’s two processors are safe and super fast.
The developers of the chip’s two processors – which both process and then compare the same data before sending onto the car’s actuators – achieved 72 trillion operations per second, well above their goal of 50 trillion operations per second, and at 68Gb per second – healthy but not ridiculous, said Bannon.
Instead of waiting for a batch of 256 units of data before processing, the AI chip processes in batches of just one – reducing latency significantly.
Also, the two processors are powered separately, meaning if one fails the car will still drive. “The probability of failing is lower than somebody losing consciousness,” said Musk.
Finally, a final security step in the chip requires a check that all data processed has been done with cryptographically signed Tesla software, so if the chip is hacked, it will not process a command issued by that hack.
The AI software learns from million of real-life drives
While Nvidia might say that this doesn’t compare to its billions of simulated hours, Musk says that simulated testing cannot replicate real-world dilemmas.
“It just does not capture the long tail of weird things that happened in the real world,” Musk said.
“If the simulation fully captured the real world well that would be proof that we are living in a simulation I think.”
All in all, it’s a pretty lengthy presentation, highlighting that Tesla strongly believes it is miles ahead of the game when it comes to full self-driving technology.
Halfway into the development of its second generation AI chip, which Musk says will be three times better than the one it revealed today and will be ready in two years, Tesla already has a huge advantage in its fleet which outnumbers other carmakers with self-driving technology by 100 to 1 according to Musk.
And that number will only increase, says Musk.
“A year from now we will have over a million cars with full self driving computer, hardware, everything.”
Bridie Schmidt is associate editor for The Driven, sister site of Renew Economy. She has been writing about electric vehicles since 2018, and has a keen interest in the role that zero-emissions transport has to play in sustainability. She has participated in podcasts such as Download This Show with Marc Fennell and Shirtloads of Science with Karl Kruszelnicki and is co-organiser of the Northern Rivers Electric Vehicle Forum. Bridie also owns a Tesla Model Y and has it available for hire on evee.com.au.