ctr_policy_ros

This ros package is intended for use with stable-baselines to load and evaluate a RL agent in the task of concentric tube robot control.

Installation and use

• Currently, the only way I have been able to get this working is by creating a python 3 virtual environment that has the required packages for stable-baselines. You can find these requirements in the requirements.txt.

• My setup has the stable-baselines cloned, to a directory and install locally from there.

• You will also need the gym-ctm-ros python package installed. You should be able to clone and install locally as with stable-baselines. Here is the link for that repo link.

• Sourcing the virtual environment, and running the nodes looks to work, as I'm only using the loading functionality of stable-baselines.

• There are two nodes in two python files in the scripts directory:

  1. ctr_sim.py: This node runs a simple simulator that acts as a simulation interface for now. It will receive actions published to "\delta_joints" and runs through one timestep to determine the new state (joint values, goal error).

  2. The ctr_sim.py then publishes the state (or obs in the code) as three seperate topics. These topics are the joint values "/joint_states", tip position "/tip_pos" and desired tip position "/desired_tip_pos".

  3. eval_agent.py: This node loads the agent and creates the policy model network. It subscribes to the "/joint_states", tip position "/tip_pos" and desired tip position "/desired_tip_pos" and creates the state required for the policy network. It then feeds it to the network, gets the output actions and publishes these under the topic "\delta_joints"