Home | Tutorials | Wiki | Issues
Ask Your Question

Revision history [back]

click to hide/show revision 1
initial version

Robot control and physics feedback

I have a machine learning project for robot controllers and I want to know if Gazebo is the right platform to use.

My project is as follows:

Its a reinforcement learning project where I need to make a robot learn how to walk. I would control the robots leg joints to make it walk (rotate the leg joints specifically). Let say I designed some stairs in my world and I want to make the robot go up the stairs. I should be able to rotate the leg joints appropriately to do so. If I do a wrong rotation, then the physics should make the robot fall. I should be able to sense the physics environment and be able to make the decision that the robot is about to fall. Then I'd take corrective measures by rotating the robot joints in some other better way.

For example, to climb stairs I need to rotate the knee joints and the thigh-torso joint. If I only rotate the thigh-torso joint without the knee joint, then the robot should start falling since that is an incorrect way of climbing stairs - and the physics simulation should take care of this and give me appropriate sensor feedback.

Can I do this sort of thing with Gazebo?