Robotics researchers from Carnegie Mellon University and UC Berkeley in Berkeley, US, have taught robotic dogs new tricks such as opening a door and kicking a ball, adding a new dimension to a robotic dog's capabilities.
Four-legged robots such as Boston Dynamics' Spot Dynamics and Unitree's Go1 have been around for years, and have arguably mastered almost every move, including walking rough terrain or climbing stairs.
New tricks
You may have difficulty teaching your dog new tricks, but the task of researchers is more difficult. Unlike real dogs, robotic dogs are not trained and cannot be tricked into learning new skills.
Instead, the researchers had to divide robot skills into two main groups, one-movement skills, and the second to manipulate objects. While we have seen multiple cases where quad robots make a single movement such as moving and climbing walls, manipulating objects was the hardest part, as the robot needs to be trained to use one of its front legs to interact with something like a door button or ball, while keeping its balance on 3 legs.
This may seem obvious and easy to the observer, but mathematically, walking and manipulating can be exactly the opposite of each other, and asking the robot to do both can put it in a math dilemma from which it cannot get out.
The researchers used a Unitree Go1 robotic dog equipped with an Intel RealSense camera to train it to walk and use its front claws to interact with simple objects. After simulations and training, the team combined all the data into one robust plan and put the dog in a series of tests to verify its capabilities, as shown in the video above.
This achievement could now open the doors to many other possibilities, as quad robots could be deployed in more realistic environments in the future.