Tell a Robot to Take a Hike, and It Might Listen: “Cassie” Takes on the Pacific Crest Trail

For most adventurers, hiking the Pacific Crest Trail would be a crowning achievement in its own right. But one Caltech professor is upping the ante: he wants to design a robot that can complete the famous trek, all with outside assistance. That’s why Dr. Aaron Ames and his team have created Cassie, the world’s first fully autonomous robot designed to navigate the harshest and most complex environments in the world.

Ames and his team at Caltech’s Center for Autonomous Systems and Technology (CAST) have set their sights on four “moonshot” challenges, each of which aims at pushing robots farther outside of the lab and into practical applications once thought to be the realm of science fiction. “We want to bring robots into the wild. Not just walking robots, but flying robots, any kind of robot you can imagine,” Ames says.

Eventually, CAST hopes to develop autonomous systems that think, act, and assist independently, expanding humanity’s ability to solve problems. That’s why Ames is eyeing the Pacific Crest Trail as the ideal proving ground for his research.

The trail, which covers 2,600 miles and 13,000 feet of elevation from Mexico to Canada, is considered one of the country’s most arduous long-distance hiking trails. And for a fully autonomous robot, the mixed terrain of the trail is a bona fide crucible. On the weeks-long journey, Cassie will have to endure everything from gravel, sand, and mud, to puddles, snow, and ice, which will push the boundaries of what autonomous robots can do. “Everyone’s seen the backflip by the Boston Dynamics robot. This approach is fantastic when you’re in a lab. But what happens if you want to do it on the beach? Things change. A lot,” Ames says. “Sand is what’s called granular media. It’s very, very hard to physically model, to have mathematical representations of it.”

Ames sees these kinds of environments occupying the intersection of dynamics and machine learning, two dueling areas of research in modern robotics. “There are people that think robots should learn everything, absolutely everything. I think that’s the wrong way,” Ames says. “Our human body is not just a brain. We have a spinal cord which has pattern generators in them that are fundamental for locomotion, for example.”

“When you’re walking down a hallway on your cell phone, you don’t have to think. And when you get to a hallway intersection or a set of stairs, you’re going to look up. So this is where a neural network would play a role or some sort of learning algorithm.”

Combining “neck down” robotics with “neck up” machine learning will eventually lay the groundwork for human-assisted locomotion. Applied in this way, “Cassie’s children,” as Ames calls the robot’s future iterations, could have deep implications for the rehabilitation of paraplegics and amputees. “We’re implementing our methods for walking robots onto an exoskeleton. We recently did some clinical tests where paraplegics were able to walk dynamically without crutches using an exoskeleton,” says Ames.

So Cassie’s trek represents much more than a mere hike in the woods. Ames’s research represents an opportunity to change how humans interact with the world. “If we’re successful, eventually we can have a paraplegic walking the Pacific Crest Trail. We’re really at the precipice of getting robots out of the lab into the wild, doing some truly dynamic things.”

__________________________________________________________________________________