Unlike existing robots on the market, such as those from Boston Dynamics Place, which moves using internal maps, this robot only uses cameras to guide its movements in the wild, says Ashish Kumar, a graduate student at UC Berkeley, who is one of the authors of a paper describing work ; it must be presented to Robot Learning Conference next month. Other attempts to use cues from cameras to guide the robot’s movement were limited to flat ground, but they managed to get their robot to climb stairs, climb over rocks, and jump over gaps. .
The four-legged robot is first trained to move around different environments in a simulator, so it has a general idea of what walking in a park or going up and down stairs is like. When deployed in the real world, visuals from a single camera on the front of the robot guide its movement. The robot learns to adjust its gait to navigate things like stairs and rough terrain using reinforcement learning, an AI technique that allows systems to improve through trial and error.
Removing the need for an internal board makes the robot more robust because it’s no longer limited by potential errors in a board, says Carnegie Mellon assistant professor Deepak Pathak, who was on the team.
It’s extremely difficult for a robot to translate raw pixels from a camera into the kind of precise, balanced motion needed to navigate its environment, says Google researcher Jie Tan, who was not involved in the study. He says it’s the first time he’s seen a small, low-cost robot exhibit such impressive mobility.
The team achieved a “breakthrough in robot learning and autonomy,” says Guanya Shi, a researcher at the University of Washington who studies machine learning and robotic control, who also did not participate. looking.
Akshara Rai, a researcher at Facebook AI Research who works on machine learning and robotics, and was not involved in this work, agrees.
“This work is a promising step towards building such perceptive-legged robots and deploying them in the wild,” Rai says.
However, while teamwork is helpful in improving the way the robot walks, it won’t help the robot know where to go ahead of time, Rai says. “Navigation is important for deploying robots in the real world,” she says.
More work is needed before the robot dog can walk around the parks or fetch things around the house. While the robot can understand depth through its front-facing camera, it can’t cope with situations like slippery ground or tall grass, Tan says; it could get into puddles or get stuck in mud.