top of page
Search
Parag Batavia

Where you Should be Going, You Don't Need Roads

At Neya, I made an early decision to focus on off-road autonomy instead of on-road / self-driving cars. I joke that this was because I really didn’t want to someday get acquired for a billion dollars (purely in hindsight - there’s some truth to that – I love my life and it seems like having that much money would change it irrevocably).


But this was back in 2010-2011, when it wasn’t as obvious that there was soon to be a revolution (in funding, hype, and capabilities) in self-driving cars. My technical background and prior work at CMU should have pushed me more towards on-road applications. But, most of the available opportunities for autonomy development was with the DoD, and they care a lot about operating off-road.


The off-road market is not strictly a DoD application, though. There are categorical differences between on- and off-road autonomy from a business perspective that will lead to substantial growth in off-road autonomy commercial applications.


Risk vs Reward – What it Always Comes Down to


The applications for self-driving cars are substantial, but by definition are limited to on-road and parking lot environments: urban and suburban mobility, on highways, city streets, semi-rural and rural roads. Off-road autonomy has an extremely diverse set of application areas, and this greatly affects the tech stack (a topic for another time) and means there are nearly as many technical solutions available as there are problem areas. The typical high-level application areas for off-road autonomy include military applications, mining, agriculture, and construction. But, each of those areas has dozens of smaller (but still huge) addressable markets.


Progress in on-road autonomy is moving at a much faster pace than off-road for one very simple reason: funding. The dollars spent on on-road applications are orders of magnitude higher than off-road. We are now seeing off-road autonomy funding from the government in the high $10s of millions to low $100s of millions per year, and VC funding is catching up – companies focusing on automated construction are raising $10s of millions in Series A and B. That pales in comparison to the billions of dollars being spent on on-road autonomy.


That is, of course, a function of the ultimate potential return and business opportunity. However, I would claim that there are a lot more opportunities for early wins in off-road applications compared to on-road, with a high ROI that can be achieved in a shorter time period.


A Highly Segmentable Market Means More Opportunities and Niches



While market segmentation is a concept typically applied to B2C products (based on demographics, geography, etc), it also applies to B2B – and segmenting based on customer needs is a common method.


There is a potential for automation in almost any application or operation where something needs to be transported or moved. In large site commercial construction alone, you’ll see a wide variety of machine types: dozers, excavators, backhoes, graders, trenchers, loaders, dump trucks, etc. Each piece of equipment is purpose-built for a specific set of operations.


Each operation has the potential to be automated, without requiring the entire construction site to be automated. This means, you can create a technology, and potentially a company, focused solely on (for instance) automated trenching. Given that 9 out of 10 US contractors are reporting shortages in skilled labor, and that construction overall is a $1.23 Trillion annual market in the US alone, smart market segmentation can lead to highly profitable businesses.


And – this is just one example of a broader market. The same is true for military applications (the DoD is getting very serious about integrating autonomy and semi-autonomy with dismounted and mounted personnel), mining applications (open pit and underground mines), and agriculture (large farms, mid-sized farms, automated greenhouses). The ability to highly segment this market, compared to driverless car opportunities, means that it’s possible to build scalable, profitable businesses without requiring a billion dollar investment as table stakes.


Cost (In-)Sensitivity Means You Can Solve Hard Problems With Expensive Hardware


The off-road market is primarily B2B, whereas self-driving car companies, even when posed as B2B, are largely governed by B2C constraints. The cost sensitivity of the automotive industry is legendary. Anyone who has worked for a tier 1 auto supplier will have horror stories of getting beaten up over a couple of pennies of cost.


Elon Musk has recently created (yet again) controversy by claiming that LADARs for self-driving vehicles are a horrible idea. While I vehemently disagree with the notion that purely vision-based approaches are ready for deployment, at scale, with the level of safety necessary, he’s right about one thing: LADARs are a cost driver that will limit the deployment of self-driving cars (even when the software is ready) until they come down in price.


Now, the automotive industry is great at driving cost down by scaling complex electromechanical systems. But it’ll take a while. This means that for the “real” self-driving future to come true, sensors will have to be dirt cheap, so they can be integrated and used on a $15,000 entry-level car. That is not going to be fast. And it’ll be hard for companies to realize gains until that happens.


[I do understand that part of the business model of most self-driving companies is to change the actual nature of transportation, and to increase ride share and increase car utilization (therefore increasing ROI per car), but that is a societal / generational change that will take a lot of time, and still likely only be possible in moderately dense urban environments. Not to say this won’t happen, but it won’t be fast.]


A $40,000 Wheel (for a $5M Loader) vs a $40,000 Tesla. Which can Afford More Autonomy?

Contrast that with the mining industry. Caterpillar has a fleet of automated 793F mining trucks that they sell to mining companies. While the cost of a 793F isn’t public, estimates are approximately $5M per truck. Each tire costs over $40K – about as much as a decent mid-sized (non-autonomous) sedan, or a semi-autonomous Tesla Model 3.


So, I’m pretty sure an autonomy system on a 793F can afford to have a LADAR – or 10 if necessary. And, the acquisition of an autonomous mining vehicle is a capital investment, so the ROI is analyzed much differently than a consumer-facing autonomous car. And while I don’t know what level of investment was needed in order to automate the 793F, it’s highly unlikely that it was billions of dollars.


If you are considering creating an autonomy company, the ROI potential for autonomy in complex construction and mining application, combined with the (relative) price insensitivity of those industries, combined with the ability to solve some challenging problems using expensive sensors, means that it wouldn’t be a bad bet.


Controlled Environments Means Fewer Safety-Related Edge Cases



Safety is paramount in all automation applications, from factory robots to self-driving cars. Any autonomous vehicle operating in military environment, construction site, or mining site will have to be designed from the ground up to consider safety at all levels and undergo rigorous testing and validation. There’s no short circuit around this.


However, industrial applications typically take place in controlled environments that just doesn’t exist in the commercial self-driving space. Self-driving companies do attempt this, by doing pilot trials and restricted operations in limited geo-fenced, well-mapped environments (campuses, military bases, suburban neighborhoods) or by restricting speed in applications like food delivery. But, even in these cases, the environment isn’t fully controlled, and the unexpected can happen.


Contrast that with an access controlled open pit mine. You will not find children, pets, bicycles, or pedestrians in an open pit mine in the middle of Australia. These sites are highly access controlled, and more importantly, everyone who has access can be trained in how to work in the vicinity of autonomous vehicles. This type of education is critical, since they don’t behave deterministically in all cases.


The opposite is not possible – in a mixed manual driving / self-driving environment (which is all that will be possible for decades), you can’t train all the manual drivers on how to co-exist with autonomous vehicles. That means the self-driving cars need to behave as close to how we’d expect a human to drive as possible (I’ve heard plenty of anecdotes about this from self-driving car execs that I can’t repeat).


So – in most off-road environments, there is at least some ability to control the environment, and people can be trained how to work with the machines. This reduces the friction (and cost) for adoption and improves the ROI argument.


Summary – It’s All About B2B vs B2C



There is a strong case to be made for off-road autonomy applications to a broad range of industrial applications. I’m certainly not the first to make this case. And, this is a typical B2B vs. B2C argument – B2B typically offers lower risk and higher margin opportunities but may not reach the extreme scale that a really successful B2C offering may provider.


There is room for both, but to date, the focus and most of the dollars have been spent on the B2C application, and there are substantial opportunities in B2B if you are willing to work on things that may seem (in comparison) less sexy than self-driving cars for the masses.

118 views0 comments

Recent Posts

See All

The Next Step

In 2009, I decided the time was right for me to start my first company. I did this having spent years learning about startup life and...

Comments


bottom of page