A machine learning model worked out how to keep the robot stable on three legs while it uses one leg to open doors.

Philip Arm, Mayank Mittal, Hendrik Kolnbach, and Marco Hutter/Robotic Systems Lab

A robot dog can use a leg. Open the doorspress the button and lift the rucksack while balancing on its other three legs.

Four-legged robots like Spot, star of Boston Dynamics’ viral videostypically requires an arm attached to its body to open doors or lift objects, but this can add significant weight and make it difficult for the robot to squeeze into tight spaces.

Flip Arm At ETH Zurich in Switzerland, and colleagues used a machine learning model to teach an off-the-shelf robotic dog to use one of its legs to perform tasks while standing still or with the other three legs. Move.

“We can’t do everything with the legs that we can do with the arm — right now, one hand is too specialized. But the real thing is to do it for applications where you have mass constraints. Or we don’t want the added complexity, like for space exploration where every kilogram of such a robot counts,” says Arm.

To train the dog — an ANYmal robot built by ANYbotics — Arm and his team tasked a machine learning model with one of the robot’s legs to find a specific point in space. The model then worked out on its own how to control the remaining three legs and balance the robot while standing or walking.

Arm and his team could then remotely control the robot to perform movements such as picking up a bag and placing it in a bin, or collecting rocks. Although the robot can currently only perform these tasks while being guided by a person, Arm hopes that future improvements will allow the dog to autonomously manipulate objects with its leg.