Off-Road Brawn with AI Brains

The phrase “where the rubber meets the road” has long been used to describe the point where theory is tested in the real world. For researchers in WPI’s Autonomous Vehicle Mobility Institute (AVMI), the expression is especially apt. They are working on technology for a new generation of autonomous off-road vehicles capable of deftly traversing the roughest terrain to perform all manner of tasks, from engaging in combat to exploring other worlds.

A critical element of this work involves the way a vehicle’s wheels engage with the ground beneath them, and how to control—and even anticipate—that interaction to ensure the vehicle stays in motion and on course, while maximizing safety and minimizing energy consumption. To meet that challenge, the research team draws on a host of advanced engineering and design methodologies and devices, including cutting-edge computer modeling and simulation, artificial intelligence, and the latest in sensor and computing technology.

Moradi, who earned his doctorate in civil engineering at the University of Alabama at Birmingham (UAB), worked in industry for nearly two decades before joining the faculty at UAB, where he served as director of research in the School of Engineering and director of the Engineering and Innovative Technology Development (EITD) center.

The two engineers began their collaboration after Vantsevich took a position as professor of mechanical engineering at UAB more than a decade ago and began exploring the idea of establishing a research center dedicated to autonomous off-road vehicles. Moradi, as head
of EIDT, led a team of nearly 40 engineers and quality managers who had, under contract from NASA, developed technology for space exploration, including a series of advanced ultra-cold freezers, many of which are still in service aboard the International Space Station. “When Dr. Vantsevich came up with his idea,” Moradi recalls, “I said, ‘Why don’t we combine forces? Then you won’t have to build your own group. We can collaborate and build this thing together.’”

AVMI’s mission focuses on two interrelated quests. The first is to develop vehicles and vehicle systems specifically designed for the off-road setting. When vehicles—whether cars, tractors, or tanks—leave the road behind, they enter an unpredictable and far more complex environment. “There are no traffic lights, no lanes, no smooth asphalt or concrete surfaces,” Moradi says. “They have to travel on rough terrains: snow, ice, mud, sand, and so on.”

More than simply conquering these tough conditions, the vehicles AVMI helps develop must clear a high operational bar. “The focus of our research,” says Vantsevich, “is terrain mobility, maneuverability, and energy efficiency, and how those factors intersect with survivability.” In other words, no matter what the terrain, the vehicles must not get stuck, they must not wander from their planned courses, and they must get the maximum mileage from their batteries. (While it also works on mechanical drivetrains, AVMI’s specialty is vehicles driven by electric motors.) On top of this, the vehicles, and any human occupants or other payloads, must live to drive another day.

Adding to the complexity of AVMI’s mission is the pursuit of intelligent autonomy. It was said of dancer Ginger Rogers that she did everything Fred Astaire did—only backwards, and in high heels. Likewise, an autonomous vehicle must do everything any other vehicle can do, but without the benefit of a human driver’s brain, sensory system, and experience. It must sense the environment, anticipate terrain changes, make decisions, and respond appropriately. That’s tough enough for a vehicle that never leaves the relatively predictable confines of roads and highways. It is exponentially more difficult in the untamed expanse of the off-road world.

“When an off-road, autonomous vehicle is moving,” Vantsevich says, “it should know what kind of terrain is under the wheels. So we need to have special sensors, like lidar, radar, cameras, and interior sensors that monitor the behavior of the vehicle and its systems.”

More than sense the terrain, the vehicle must anticipate it to make fine adjustments, millisecond by millisecond, in the position and movement of each wheel to ensure that the vehicle keeps going and stays on course, no matter what kind of muck or mire it may encounter. “In these multi-wheeled vehicles, we control each wheel individually with electric motors,” he says. “Even before a wheel starts spinning, we must already have started to control it.”

Vantsevich and Moradi believe the autonomous technology they are developing can be deployed in a wide range of vehicles, including earth movers, tractors and other farm vehicles, and planetary rovers. (They are in early talks with NASA’s Ames Research Center in California.) Noting that the autonomous aspects of AVMI’s work aligns well with President Grace Wang’s vision for the future of AI at WPI, Vantsevich and Moradi say they plan to collaborate with many of WPI’s schools and departments, including other researchers who focus on AI.

“GVSC is the center of excellence for all types of vehicles for the U.S. Army,” Vantsevich says. “The next generation of military vehicles, manned and unmanned, will have GVSC’s signature on them.” Adds Moradi, “They have a strategy for that next generation. They want to go autonomous: fewer people on the battlefield, less danger to human life. That’s their vision, and survivability, energy efficiency, maneuverability, and agility are all important pathways for getting there.”

To accelerate progress toward that goal, GVSC created the Ground Vehicle Modeling and Simulation Alliance (GVMSA), a network of university research centers that focus on digital engineering, virtual prototyping, and the modeling and simulation of ground vehicle systems. Funding for university research on ground vehicle systems flows from GVSC through a multi-university alliance known as the Automotive Research Center, which is anchored by the University of Michigan.

In addition to this core group of university labs, two academic research groups have special roles to play in achieving the Army’s vision for GVCS, in collaboration with industry: a center on virtual prototyping at Clemson University and WPI’s AVMI. “They consider these two universities to be important links to applied research and to industry,” Vantsevich says. “This makes AVMI a key element in the alliance.”

AVMI pursues its mission in a suite of modern offices and laboratories on the third floor of a former mill building on Prescott Street in Worcester, just a stone’s throw from WPI’s Gateway Park research and development complex. There, a team that includes four research engineers, scientists, and associates; a PhD student; and 11 graduate students and undergraduates—as well as multiple teams working on Major Qualifying Projects—explores advances in autonomous off-road vehicles, both physically and virtually. In addition to the working relationships the institute is forging with WPI faculty in such areas as computer science, engineering, mathematical sciences, and robotics, it continues to expand its collaboration with private industry partners, including Volvo Construction Equipment, Nikola Corporation, Boston Engineering, Waltonen Engineering, and Drive System Design.

On the physical side is a testbed vehicle (about the size of a large microwave oven) built from scratch over the course of four years at a cost of more than $100,000, and continually updated and outfitted with the latest technology. “This vehicle is equipped with the most up-to-date sensors in the world,” says Siyuan (Hunter) Zhang, PhD, research engineer and lab manager for AVMI. “So, we have lidar, stereo cameras, GPS, and IMU [inertial measurement units], along with the most advanced computers and controllers. We can test the hardware and control algorithms we develop using this vehicle, including the algorithms that enable us to control each wheel independently to increase mobility and maneuverability.”

While physical tests are vital to AVMI’s work, much of its research is done virtually, within the confines of high-performance computers. That includes the development of computational models that can simulate every aspect of a vehicle and the environment it must navigate. “Let’s say you have a vehicle moving in a virtual environment,” Moradi says. “The mathematical descriptions of its motion and its interaction with the environment are all physics-based. That means that the vehicle is dynamically interacting with the terrain, which is deformable. The vehicle deflects the terrain, and all the deflections occur in real time as the vehicle is moving.”

As new vehicles and systems progress from concepts to engineering designs to prototypes, models and simulations enable the AVMI team to test dozens of iterations, in myriad virtual environments, quickly and accurately, without having to build physical devices. “Before you put something into a real vehicle, you can test it here, virtually.”

“The results of computational modeling also help train machine learning algorithms and other kinds of artificial intelligence applications that will be critical to the development of autonomous vehicles and systems that can make accurate predictions about the upcoming terrain and respond quickly enough to avoid getting into trouble,” Moradi says. “Training the AI for an unmanned off-road vehicle will be far more difficult than training for a road vehicle.

“AI algorithms will also help us develop robust, secure systems that can quickly detect if a sensor is sending out an anomalous signal or if an adversary has compromised your systems and distorted the sensor information. So, an autonomous system must be intelligent enough to understand when a signal is wrong, to recover the correct signal and discard the wrong one, and to protect sensors from being compromised.”

Instead of ordinary walls, the lab will have walls covered in LCD screens that will display a 360-degree view of a simulated vehicle. Researchers will be able to literally step into a simulation and see the operation of a vehicle and its interaction with its environment in a way that will be more realistic than any other type of immersive technology. The lab will also make it possible for researchers in other locations to join in the immersive experience.

“If you are somewhere else in the world,” Moradi says, “you can put on VR goggles and feel as if you are in the lab, participating in the simulations. This will enable us to make AVMI an extension of the Army’s GVSC. People there will be able to run experiments in our lab.”

That kind of connectivity and virtual collaboration will also enhance other collaborations the institute is building or hopes to establish, including those with teams at the University of Hawaii at Manoa, Texas A&M University, and several NATO countries.

The purpose of models and simulations is to refine and perfect vehicle designs, but AVMI’s job does not end there. “The Army likes to work with us because we are not just doing basic research,” Moradi says. “We go further. We go to applied research, then to conceptual
and engineering design. Then we use the systems process to build prototypes and, eventually, commercial systems—what the government refers to TRL 9, or Technology Readiness Level 9, the highest level.”

“This is what Dr. Moradi actually did at UAB,” Vantsevich says, “and it is what I did throughout my professional career working on off-road trucks, farm tractors, and construction equipment that went into production. This is what GVCS and our corporate partners wants from us, to do the applied research, the engineering design and prototyping, and then to go to industry to build the hardware that will go to market.”

And as those advanced autonomous off-road vehicles come off the assembly line and head out to do real work in the real world, it will be research and innovative ideas from AVMI that keeps them rolling along.