A machine learning model figured out how to keep the robot stable on three legs while opening a door with one leg.
Philip Arm, Mayank Mittal, Hendrik Kolvenbach, Marco Hutter/Robot Systems Laboratory
The robot dog can open doors, press buttons, and pick up backpacks with one leg while balancing on its other three legs.
Quadruped robots like Spot, the star of Boston Dynamics' viral video, typically require arms attached to their bodies to open doors or lift objects, which adds significantly to their weight. This can make it difficult for the robot to maneuver through tight spaces. .
philip arm Researchers at ETH Zurich in Switzerland used a machine learning model to teach an off-the-shelf robotic dog to perform tasks using one of its legs while remaining stationary or moving with the other three. I taught you to do it.
“We can't do everything with our legs that we can do with our arms. We're much more dexterous with our hands at the moment. But what's really important is making this work in applications where there are mass constraints, or in robots. “The idea is to make this work in applications where you don’t want the added complexity, such as space exploration, where every kilogram counts,” Arm says.
To train the dog, the ANYmal robot from ANYbotics, Arm and his team gave the machine learning model the goal of finding a specific point in space on one of the robot's legs. The model then took control of his remaining three legs and independently worked out how to keep the robot balanced when standing and walking.
Arm and his team can now remotely control the robot to perform actions such as picking up backpacks and putting them in boxes, or collecting rocks. Currently, the robot can only perform these tasks when controlled by a human, but Arm hopes future improvements will allow the dog to autonomously manipulate objects with its paws.
topic:
Source: www.newscientist.com
