Watch robotic dog perform new tricks you haven't seen before

Teaching the robotic dog new tricks

If you have had a tough time teaching your dog new tricks, the task for the researchers was even more difficult. Unlike real dogs, robotic dogs do not respond to treats and cannot be fooled into learning new skills.

Instead, the researchers had to compartmentalize the robot's skills into two major sets, one locomotion, and the other being manipulation (of objects, of course). While we have seen multiple instances of quadruped robots moving and climbing on walls, the manipulation is the trickier part.

In this, the robot needs to be trained to use one of its front paws to interact with an object such as a button door or a ball, while maintaining its balance on the three. This might seem straightforward for an observer but computationally, locomotion and manipulation can be quite the opposite of each other, and asking the robot to do both can set it into a computational tailspin from which it cannot get out.

The researchers used a Unitree Go1 robotic dog with an Intel RealSense camera to train it to walk and use its front paws to interact with simple objects. After training the robot in multiple simulations, the team combined all the data into one robust plan and put the dog through a series of tests to verify its capabilities, as seen in the video above.

This achievement can now open doors to many more possibilities where quadruped robots can be deployed in much more realistic environments in the future. It is definitely a better way to deploy a robotic dog than simply mounting a gun on its back to fire at will at trespassers on your property.

The ability of the dog to press wheelchair buttons can be incorporated into robotic companion dogs in the future. Any more ideas for newer tricks anyone?