Allen is a Ph.D. candidate at Stanford University, where he specializes in training robots to dodge obstacles at high speeds. He does most of his research on small quadcopters, as they’re one of the most adaptable, dynamic, and controllable platforms to develop collision-avoidance technology. Plus, they’re pretty damn cool.
“[Quadcopter drones] are a great research platform in that they’re complex enough that they add all these challenges that researchers like to see,” Allen told Inverse. “They’re a good mix of challenge and usability.”
But quadcopters are just the building blocks of Allen’s research, which could be groundbreaking in several robotics industries.
“The research is… really applicable to any sort of robotic systems especially when those robotic systems are sort of fast moving, or they have behaviors they have to obey, such as cars,” Allen said.
Autonomous or self-driving cars could be ubiquitous in the near-future, but one of the biggest challenges in artificial-intelligence drivers is getting them to sense and react appropriately to fast-moving, external obstacles in varied circumstances. Allen is demonstrating his new technology in a simple, low-tech way (by swatting at it with a sword), but he says the high-level capabilities of his technology are “less flashy” than the sword-dodging.
At a higher level, the autonomous dodging software (technically called “real-time kinodynamic motion planning”) can help vehicles or equipment navigate from point to point on a plotted course, while responding to obstacles as they pop up unexpectedly. That means a delivery drone or self-driving car could swerve to avoid an obstacle and then automatically re-plot its course to find its way back on track. Allen says it could also help autonomous boats, heavy robotic cranes, or even space ships react to minute variables during complicated procedures like picking up objects or docking with space stations.
“I came into this research wanting to address a larger problem,” Allen said. Even though A.I. technology is rapidly advancing, teaching robots to react to huge amounts of variables as quickly as humans can is the next step to true autonomy. In the mean time, let’s just hope and pray nobody combines Allen’s technology with the crazy chainsaw-drone, because based on the video below, Allen’s fencing sword wouldn’t stand a chance.