The recent crop of early-stage startups – along with several recent VC investments – illustrates an emerging area of interest in the autonomous vehicle technology sector. Unlike corporations bringing robotics to city streets, these startups are taking their technology to the field.
Two recent entrants – based in Seattle Land artificial intelligence and based in New Brunswick Potential — are ready to achieve a first-mover advantage in this autonomy segment.
While these startups use their technology in alternative ways, Overland AI and Potential share common ground. Every startup’s founders consider they’ve cracked the code on one of the tougher automated driving applications by creating software that doesn’t rely on some of the primary pillars of testing and implementation – equivalent to detailed maps, large swaths of coaching data, and the ability to make use of distant assistance.
The US Department of Defense and enterprise capitalists are taking notice.
Overland AI, which is developing an autonomous guidance system designed for military operations equivalent to reconnaissance, surveillance and delivery of electronic warfare packages, was awarded in April to $18.6 million from the U.S. Army Defense Innovation Unit. The funding will probably be used to build a prototype autonomous software stack for the Robotic Combat Vehicle (RCV) program over the next two years.
The 2022 startup raised a $10 million seed round led by Point72 Ventures this week. According to CEO and founder Byron Boots, the funds will probably be used to expand the Overland team and further develop OverDrive, the company’s autonomy suite.
Meanwhile, Potential, which makes advanced driver assistance systems (ADAS) that enable off-road vehicles, underground mining vehicles and passenger cars to navigate off-road, has raised a C$2 million (~$1.5 million) seed round led by Brightspark Ventures, a Canadian early-stage VC fund. This brings Potential’s total funding to C$8.5 million (~$6.2 million). The startup has spent the last six years developing its technology and is currently running several pilot projects spanning motorsports, motorcycles and automotive.
Possibility of off-road driving
Potential and Overland AI aren’t the only corporations attempting to apply autonomous vehicle technology to areas beyond public streets. The costly pursuit of economic robotxi and autonomous truck operations has thwarted dozens of startups over the past few years. After their closure, a recent group of startups emerged, equivalent to Polymath Robotics, Forterra, Pronto.ai, Bear Robotics and Outrider, with more established ambitions: applying AV technology to warehouses, mining, industrial and field environments.
“We are absolutely investing capital in field autonomy,” Alexei Andreev, managing director of Autotech Ventures, told TechCrunch. “In fact, if anything, we stay away from highway autonomy and focus entirely on off-road autonomy.”
Most of the off-road corporations in which Autotech Ventures invests today are in the agriculture and construction sectors – producing products equivalent to autonomous mining vehicles, forklifts and tractors. For these sectors, Andreev says the idea is to deal with labor shortages while increasing productivity and making farms and construction sites safer.
“And if you remove people, you immediately get a reduction in your insurance premiums. So the return on investment in vertical applications is now significant,” Andreev said.
Another result: autonomy in the field found a friend in defense.
Overland AI: From DARPA to Seed Funding
When it involves off-road driving automation, the US military may very well be a great customer. After all, autonomous vehicles were created as a part of a DARPA project, says Jeff Peters, partner at Ibex Investors. DARPA (Defense Advanced Research Projects Agency) is an agency of the United States Department of Defense responsible for developing technology for military purposes.
“The hype around AV has moved much of the industry toward larger potential commercial applications, but the Department of Defense projects have not stopped,” Peters told TechCrunch by email, noting that autonomous mining startup SafeAI and autonomous trucking startup Kodiak Robotics also applied for defense grants. “I think AV companies (those that still exist) will pursue DoD projects because they offer large, non-dilutive financing in the interim period before commercial operations.”
Overland AI is the latest by-product of the DARPA program. Boots, a professor of machine learning at the University of Washington and founding father of the Robot Learning Laboratory in the university’s school of computer science and engineering, has a long history of collaboration with the U.S. Army Research Laboratory and DARPA.
Overland was born out of research by Boots and a team involved in DARPA’s RACER (Robotic Autonomy in Complex Environments with Resiliency) program, which goals to develop autonomous vehicles that can cope with difficult terrain.
The program is still implemented. Overland, which incorporates high-tech veterans from Google, Nvidia, Apple, Waymo, Aurora, Embark and Argo, in addition to software engineers who have worked on mission-critical solutions at SpaceX, RTX and the U.S. Army, was recently chosen to maneuver to second phase.
“The basic idea is that almost every ground vehicle used by the military today has a human in it,” Boots said in a video interview with TechCrunch. “You can imagine being able to just pull the person out of the vehicle, which provides safety and tactical benefits.”
Removing a person means vehicles must autonomously navigate complex off-road terrain using only on-board sensors (mostly cameras, in keeping with Boots) and computation, without relying on maps, GPS or distant operators. This means that Overland’s software must understand the ground geometry – including vegetation and mud – at every turn and its impact on vehicle dynamics.
“The terrain affects how the vehicle moves,” Boots said.
Overland’s technology “basically takes data from sensors and creates a representation of the terrain on the fly,” Boots explained. The vehicle then uses this digital representation “plus the destination it is trying to reach, which may be several kilometers away, to find its way across the terrain to that destination.”
“Part of the benefit of having an autonomous system is that when the system receives a task, and if you lose the communications link to the ground vehicle, it will continue to the destination and attempt to complete the task until the communications link is restored,” But said.
Most modern road vehicles rely on a telecommunications link to enable distant assistance, in part because the risk to other road users is higher. That’s why you may see Waymo and Cruise robots bricked up on the streets of San Francisco, waiting for a distant operator to nudge them when they stop driving to satisfy minimum safety requirements.
“Military ground systems often need to operate in unstructured, dynamic terrain. We believe that self-driving technology built for well-defined streets and closed areas will struggle there and that a very strong team is needed to ensure operationally adequate ground autonomy in such environments,” Chris Morales, partner in the defense technology team at Point72 Ventures, told TechCrunch.
Potential potential in off-road ADAS
“How do you actually enable someone who may not be a 100% experienced driver, but someone who wants to go off-road and experience more challenging conditions?” – asked Sam Poirier, CEO of Potential, in a recent interview.
Potential’s core platform, called Terrain Intelligence, uses computer vision to assist vehicles see, interpret and prepare for complex terrain and changing road surface conditions. Terrain Intelligence can read data from a single camera, slightly than relying on additional sensors equivalent to additional cameras, lidar and radar.
At its most elementary level, the ADAS Potential off-road system alerts the driver to an impassable object ahead or the need to regulate the driving position for the higher depending on the recent terrain.
“The second level is, can we instead actually help automate changes to settings that are typically driver-assisted?” Poirier said. “Most vehicles have two-wheel drive, four-wheel drive, sand mode, mud mode, and the like. Ultimately, at this stage, it’s up to the driver to switch between them… and they need to know when to use these different modes.”
The ultimate level of potential would involve leveraging existing sensor data and fine-tuning these settings and pushing the boundaries of performance.
“There are things that assistance tools can do that an individual driver – no matter how good your knowledge – cannot do on their own,” said Scott Kunselman, former Jeep chief engineer, automotive industry veteran and Potential advisor. “Stability control is a good example because independent brake control is needed to enable stability control. The driver has only one brake pedal at his disposal and activates the entire braking system simultaneously. In contrast, stability control can actuate each wheel individually and thus provide, for example, the ability to compensate for vehicle yaw.”
By the way, yaw is when a vehicle’s weight shifts from its center of gravity to the right or left, which may cause it to rotate or “fishtail.”
Potential said it is working with each Tier 1 suppliers and OEMs to license its software and integrate it directly into vehicles. Andreev suggests potentially focusing on business relationships with Tier 1 suppliers slightly than OEMs, who are less prone to take risks with a small startup.