Set up :

We will use gymnasium for our RL based learning and mujoco as the physics simulator/environment.

pip3 install mujoco pip3 install gymnasium

git clone https://github.com/WillMandil001/RL_Franka_Pushing.git

Then you need to build the custom environment (INB01014-v0)

pip3 install -e gym-INB0104

Running the current simulation:

cd src python3 gym-env-test.py

editing and adding to the simulation:

• The Gymnasium RL training script is found at /src/gym-env-test.py
• The gymnasium simulation environment is found in /gym-INB0104/envs/INB0104
• The INB0104 .xml head script is found in /environments/INB0104/Robot_C.xml

ToDo:

• we need to set up the reward to be correct.
• we need to set up the DeepRL velocity controller. (camera RGB image as input)
• Train the DeepRL policy
• implement DeepRL network on the real robot and setup real world velocity controller
• add safety control to the real world controller
• test the system and asses sim-2-real gap for the camera and the robot.
• implement Real World fine-tuning.