Interview with Josh Hartung — Co-founder and CEO of Polysync
This is one of a series of interviews with guests of Stanford’s ME302C: The Future of the Automobile — Mobility Entrepreneurship (see more here)
My conversation with Josh Hartung brought up a number of points that are rarely discussed when considering the future of transportation. First, Josh stressed just how complex the systems are on which autonomy is being built. While the industry and its avid followers tend to focus on the visible technology and progress being made, the backend tends to stay in the background. This is a part of the equation that is being overlooked and will need to act as the foundation upon which all else stands. Security and reliability in this domain will be essential. Second, I found Josh’s views regarding the “autonomous industry” to be salient. While this is not yet an “industry”, those involved have not exactly tempered expectations or the level of progress that has been made. While this is helpful from a marketing and a hype perspective, it could cause long term harm. Both consumers and regulators may begin changing their behavior when there is still so much to be accomplished. Finally, I appreciated Josh’s remarks on the automotive industry potentially adopting some characteristics of the air travel industry when it comes to safety. His avid endorsement for an open and transparent approach is a noble one. It seems unlikely this openness emerges anytime soon, however, as companies race to perfect their technology. Perhaps in the long term, we will move towards an automotive industry where safety is no longer a luxury but a normality.
When do you think we will see autonomous vehicles become mainstream?
“It’s hard to speculate on the “when.” There are some considerations to take into account when answering that question. For me, the most interesting areas are those that involve both an economic incentive and the ability to limit complexity — which makes trucking the ideal first case. It’s a massive industry and there are controllable conditions.
The hardest application is the personal autonomous car. That will most likely arrive at production last due to its naturally difficult operational envelope. The personal autonomous car has to have the ability for a global roll out.The payoff for this may not even be that great due to a smaller customer base, as well as our inability to predict what TNCs will look like in that future state. I’m not saying car ownership is going to go away, but I do think it will be reduced.”
What is the biggest obstacle to widespread adoption?
“Safety is the biggest obstacle, and that’s why we’re focusing on it. There are massive gaps in the backend systems of AVs that would be considered safe enough to put members of the public in. One of the biggest contributors is simply the sheer complexity of the system. It is so much more complex than anything that has been done before.These systems are a new form of computing. They have the processing of supercomputers, the connectivity of cell phones, and require a critical safety level of a commercial airline or better.
While we develop AVs, everyone is focusing on the application stack: perception and AI. But without a solid platform, it’s a skyscraper built on fill. What happens when the back-end of these systems fails? The likelihood of failure will continue to rise as the technology becomes increasingly complex. This makes the industry’s current approach to safety insufficient. You can’t extend today’s standards of validation and verification to what will be required on a production AV.
As a comparison, when we first looked at the cloud, we wanted mission critical systems in place. As a result, we set up mainframes. They were very expensive, but they worked. It was crucial to have the right hardware. Now, those systems have been replaced by distributed systems in order to shift load and keep resources working. Netflix did this during the AWS shutdown. The car of today is becoming a bespoke mainframe. The car of tomorrow should look more like a central computing spine made up of a number of big computers that are used for all tasks.
If you look at car safety, there is a hierarchy of time. Decisions are taking place on different magnitudes of time. The longer it takes to make a decision, the less safety critical they become. In the higher frequency decisions, you cannot tolerate latency. Therefore, you have to stay local to a car and achieve high assurance within the car. This is why automotive is big for Intel and Nvidia. It’s the last place where their products will be needed in large quantities.
Safety and regulation are not nearly as closely tied together as we think they are. Often, safety is considered only after something bad happens. Currently, there is no safety critical environment that you can deploy into AV systems today that preempt the “something bad.”. That alone makes safety the biggest untold story in this industry. ”
How do you think the dynamics play out in the industry between ride-hailing companies, OEMs, tier 1’s, the big tech companies, etc?
“In terms of market dynamics, autonomous driving is not yet an industry. Nobody can currently buy anything that is autonomous or obtain a service enabled through autonomy. As a whole, the industry is in flux. A period of transition is occurring right before we come into our own as an industry in a more mature sense. The hardest part of it all is figuring out how to stay in — that’s why a strong business model is so key.
If we think about business models in this industry, it’s important to first look to the past. The one great business model that has existed is one where you build cars and sell them to people. A lot of companies have tried to enter this market and replicate this model, but few have succeeded. Even Tesla, the most impressive example, still has challenges.
Now, as the industry evolves, there are a couple of other models emerging. One of them — our model — is selling directly into the automotive supply chain. The other is to make an end-run and try to sell directly to the customer. Think: Uber, Lyft, and even Starsky in trucking. Some of the biggest questions going forward will revolve around how companies actually figure out how to generate revenue, which is far from clear at this point.”
How are you approaching selling into supply chains?
“Our approach is subtle, making it often overlooked. Basically, we think that anyone building safety critical robotics applications should be focused on the software it’s running on. We occupy a meta-layer for all these companies. Other suppliers, like those providing AI, LIDAR, etc. are also taking on similar business models, tending to fall under the category of technology and application providers.
When it comes to safety and security, there is value in being open and demonstrating resilience and thought process. The open approach wins out because safety is not something we should be competing on. You shouldn’t be able to buy more safety just because you’re spending more money. The automotive industry has actually always operated this way, where higher quality products were safer. But this is different in air travel: one crash in that industry is bad for the whole industry. I predict this will be the same for automotive going forward. If there is a crash or an incident, people will trust the whole industry’s technology less. Safety has to be an industry-wide consideration. It’s irresponsible to think otherwise.”
What is the greatest misconception in the industry right now?
“There’s a perception right now that self-driving technology is going to completely change our lives. Everyone is hyper focused on the end-state. There is this public perception that these technologies are ready and that they are right around the corner. Nobody is abating that sentiment and some are even pushing up their timelines, saying they’ll be ready in 2018 and 2019. As a result, what you’re seeing is a dangerous regulatory reaction, where people think we need to have laws surrounding this technology figured out now. If you contrast this with other safety innovations, such as airbags, you’ll see that they go through iterations and change.
Autonomous is not solved — there are still novel approaches that have yet to be developed. We shouldn’t be limiting these changes with preemptive regulation and inflated promises. There is a danger surrounding the creation of unrealistic expectations, leaving the industry vulnerable to setbacks amid a strong negative reaction from the public after a potential first crash. It’s okay that our current systems are not on safety critical back-ends. We are taking the right steps and precautions to figure things out. These technologies are very experimental right now. We don’t know how things will unfold. So let’s take a breath and let them.
While most companies don’t yet know the level of rigor needed, they are aware of the necessity of its identification. I’m very optimistic about these technologies. Most companies are aware of these challenges and are working on them. We are seeing very few issues relative to the complexity of the challenges. Yes, the stakes are high. But I’m confident that we will come to the right solutions. It’s fine for the time being to run on experimental software and hardware as this technology develops. But in the future, the public will want something that is held to rigorous standards of safety when we step into these vehicles, or put our children in them. I have no doubt that AVs of the future achieve that level of safety.”