Tesla’s release of its v7.0 firmware last month brought with it the first of its highly anticipated autopilot software suite to the streets. With the initial launch, the masses were nervous about taking their hands off the wheel, but managed to get over their fears enough to try it out. Videos began surfacing of people turning on autopilot and their nervous but excited reactions… then too quickly moving to over-confidence bordering on just plain stupid, and even a few near misses that seemed to have been the fault of autopilot.
Several days later, there were stories of autopilot saving the day, multiple stories sharing that autopilot was improving, and not just in small ways, but very quickly, dynamically, on a day-to-day basis. Tesla shared news of an early update in the works dubbed v7.1, which promised more large features and several bug fixes. Overall, the reception was positive and truly set the high-water mark for autopilot features across the industry.
Tesla’s approach has been different than others in the industry when the specifics are torn apart, but at a high level, it’s a suite of highly responsive sensors working together in tandem to give the car the best understanding possible of what’s happening. That data, combined with the already known vehicle mechanical logic (traction control, speed, etc) is then rolled up with fixed logic and dynamic intelligence that must decipher the wide variety of inputs being presented to drive the car. Easy, right? Not on your life.
Autopilot Looking Forward
A recent article posted on sister site Gas2 shared insights from BMW’s R&D Chief Klaus Fröhlich into just how challenging developing this logic is/will be:
The software is still far less capable of making instantaneous decisions when an unexpected situation occurs than a human driver is. In an emergency, Fröhlich asks, “Do I steer into a wall? Do I steer into the human being? Do I steer into the oncoming truck? What will I do? This technology causes many headaches,” he says, “And, honestly, because it has to prevent accidents, it has to be state-of-the-art.”
He also disagrees with those who say that autonomous driving will eliminate traffic accidents. “There will be accidents with autonomous cars, it is very clear, because people behave randomly. This… also places a particular burden on the vehicle manufacturer. A child runs on to the road. As a vehicle manufacturer, I have to prove that I have done everything to account for the environment as best as possible. And we need regulations for when these accidents occur.”
Toyota the Divergent
With most of the incumbent auto manufacturers touting R&D models that “did this 10 years ago” or have this one feature built in on a tech package nobody bought, one is taking a sharp turn from the raging river that is autopilot and throwing a cool $1 billion dollars at it — Toyota. Well known to be a deviant, having challenged the masses pursuing electric vehicles as the solution to zero-emissions personal transport with its hydrogen fuel cell–powered Toyota Mirai, we should have known it would be the one to chart a new course. Toyota has turned to one of Elon Musk’s arch nemesis — AI — to solve the riddle of autonomous driving. Musk has long been publicly cautious of AI to the point that he has intentionally invested in leading AI firms to keep tabs on them from the inside.
To translate its goal of autonomy through AI, Bloomberg View shared the news of Toyota’s announcement of a $1 billion dollar investment to found its new Toyota Research Institute in the US. It did not share timing on when it foresees autonomy or even autonomous features being available in its vehicles, but it did lay out some of the high-level goals of the Institute:
3 Goals of the Toyota Research Institute
Without a firm definition of its specific intentions with Artificial Intelligence, Toyota framed its approach with 3 high-level goals, with safety being the first and foremost. Like many other car companies, Toyota would like to use AI technology in its vehicles to improve the safety of its vehicles regardless of driver actions. This sounds on the surface like Tesla’s active safety features and can be found on many high-end cars with alerts, automatic braking, etc.
Second, Toyota aims to improve accessibility with its AI-enabled vehicles, making personal transport available regardless of physical challenges. As medical technology improves our ability to sustain and extend life, improving the capability of personal transport to accommodate and enable persons with accessibility challenges is a key to maximizing their quality of life.
Finally, robotics to allow our aging population the dignity of aging while providing mobility throughout.
Serial vs Parallel Autonomy
Toyota framed up these goals against two specific application classes — serial and parallel. With serial autonomy, the car will do everything — from driving a drunk driver home (who is presumably unable to back a “backup driver” in an emergency) or a driver who chooses to sleep while the car drives home/to work/etc. This is essentially the full autonomy that everyone is aspiring to, possibly with several iterations as features are vetted and released.
Parallel autonomy is essentially active safety features that can and will dynamically monitor for hazards and respond in parallel to a human driving the car to its destination. I can imagine this evolving over time into a more robust autopilot-style suite, though this was not discussed. It starts with the barebones safety — slow and/or stop when car in front slows, respond to a car merging into the lane, alert the driver when approaching an object in front too quickly, etc, then adds on features to manage cruising on the highway, staying in the lane, dynamic speed changes… much like the Tesla autopilot we know today.
Much like the recent Tesla quarterly update call, there were 2 nuggets shared in the video that stood out to me.
- “Toyota wants to apply AI much more broadly.”
Toyota plans for the AI generated from its new center to be applied much more broadly than in Toyota vehicles or even in the automotive industry. This presents a new opportunity for Toyota to leverage a single platform and technology to drive business (see what I did there? :P). The possibilities are quite open ended at this point, painting broad swaths of industry.
- “AI can be used to schedule cars.”
Of note, Toyota shared that its AI could be used to schedule cars. The obvious connection is with Toyota cars, leveraging their integrated AI into a competitive advantage for the brand when being considered for transportation services like Uber, Lyft, etc to manage fleets and operations thereof. I’m not going to speculate like many have coming out of Tesla’s earnings update, but it did trigger me to listen closer for a bit.
With Tesla’s autopilot live in the field, Toyota is behind the 8 ball to deliver solid serial and parallel offerings… and to do so quickly. Toyota acknowledged that “we are just beginning now” at least 3 times, so the company knows it has a bear of a task ahead and the sheer magnitude of its initial investment speaks volumes about its thoughts on what’s needed to deliver and where this falls in its order of priorities.
The full press conference where Toyota broke the news on its investment in a new Toyota Research Institute in the US is on YouTube. The actual talk is only ~15 minutes and is followed by Q&A and a photo session (yup).