An engineer from Stanford University, Marco Pavone, is developing technologies to help robots adapt to unknown and changing environments.
Pavone is an assistant professor of aeronautics and astronautics, who has worked in robotics at NASA's Jet Propulsion Laboratory. He and his students are devising algorithms to help robots make decisions on their own, within fractions of a second. The immediate application is to help robots navigate independently to bring space debris out of orbit, deliver tools to astronauts, and grasp spinning, speeding objects out of the vacuum of space.
When grabbing objects in space, there is no margin for error. Accidentally bumping an object in space could make recovering it next to impossible.
"In space when you approach an object, if you’re not super careful in grasping it at the moment you contact it, the object will float away from you," Pavone said.
To solve grasping problems, Pavone teamed up with Mark Cutkosky, a professor of mechanical engineering, who has spent the last decade perfecting gecko-inspired adhesives. The gecko grippers allow for a gentle approach and a simple touch to "grasp" an object, allowing easy capture and release of spinning, unwieldy space debris.
Another piece of the puzzle involved is the delicate navigation required for grasping in space.
"You have to operate in close proximity to other objects: spacecraft or debris or any object you might have in space," Pavone said. "That requires advanced decision-making capabilities."
Pavone and his collaborators designed algorithms that allow space robots to autonomously react to variable conditions and efficiently grab space objects with their gecko-grippers. The resulting robot can move and grab in real time, updating its decisions at a rate of several thousand times a second.
These robotic decision making techniques could also be used on Earth, where they could be used to improve self-driving cars and drones.
For drones, in particular, which have to work in environments navigating at high speed in proximity to buildings, people, and other flying objects, this decision making could be very useful.
To this end, the engineers are working on "perception-aware planning" which allow drones to not only consider fast routes, but also "see" their surroundings and better estimate where they are. They are also hoping to expand this work to handle interactions with humans, which is essential for autonomous systems like drones and self-driving cars.