Moveit VS RL: Motion planning and Path planning

[naa1824]

What I understand is, in manipulation applications, Moveit can create the path planning and Motion planning to generate the motion.
But another approach to do so is to use machine learning: Reinforcement Learning to replace Moveit. so the robot will generate the motion based on the trained policy.

How does this Repo merging both together?
https://github.com/AndrejOrsula/drl_grasping

Can you teach me what is the role of each one in this project?

[AndrejOrsula]

Hello @naa1824,

That would depend on the action space of your RL agent.

• For joint space commands (e.g., target joint position actions), you would not need MoveIt as you could simply plug your agent's action output into the underlying joint controllers.
• For task space commands (e.g. relative end-effector displacement), you need an intermediate layer between the output of your high-level agent and the low-level joint controllers. In this case, MoveIt can be leveraged to provide a collision-free path for each action of the agent (e.g. move end-effector 5 cm in the positive X direction using IK), which provides control signals for the joints. You can use other controllers for the same purpose, e.g. Differential IK (MoveIt Servo), Operational Space Control, ... MoveIt is one of the many options to achieve the same goal (it actually has some limitations for RL, especially during training where it would be difficult to parallelize).

This project leverages the second approach in order to have a robot-agnostic action space that does not depend on the number of joints. For some tasks, this also simplifies the problem due to the reduced state space that the agent needs to explore prior to finding rewarding states. On the other hand, it also reduces the potential capabilities of the agent as it does not have full control of its kinematics.

[naa1824]

I see
so in simple word, RL is controlling Moveit. The outputl of the RL model, is the input of Moveit

Thank you so much

[naa1824]

I believe you should start teaching at universities if you haven't started already

Wish you all the best