Product: [Jumping Sumo]
SDK version: 3
I’ve created a Gazebo simulation of the Parrot Jumping Sumo which is quite close to a real Sumo. I’ve done a video that shows a side by side demo of the movements of a real sumo being recorded with ROSBAG and then being fed into the Gazebo simulation on the right:
The goal of creating the simulation is to use reinforcement learning to teach a sumo to do things like climb stairs in Gazebo (similar to reinforcement learning in https://github.com/erlerobot/gym-gazebo with a turtlebot) and then apply it to a real Sumo and hopefully get a real sumo climbing stairs. Unfortunately my jumping mechanism on the Sumo stopped working so will have to teach the sumo to do something other than climb stairs. I’ll put up a github project with my work with reinforcement learning if I can teach the sumo to do something worthwhile.
The source of the project is at:
If anyone at Parrot is interested in my work I am keen to do work with reinforcement learning, deep learning etc with drones.