Evaluate in the CPM Lab

This guide explains how to run a trained SigmaRL policy through the CPM Lab ROS 2 workflow.

Before You Start

This page assumes that:

  1. SigmaRL has already been installed inside the CPM Lab workspace as described in Install SigmaRL within CPM Lab.

  2. You have a trained SigmaRL policy folder available, as described in Evaluate in SigmaRL.

If these prerequisites are not satisfied yet, first complete the linked installation and SigmaRL evaluation guides.

Prepare the CPM Lab Simulation

Before starting the SigmaRL controller, prepare the CPM Lab simulation setup. Follow the External Motions Planner workflow until the experiment setup screen.

When selecting the map, choose Lab Full. The external motions planner example uses Lab Outer Circle, but the SigmaRL CPM Lab evaluation should use the full lab map.

Important

If the Control Center offers a built-in high-level controller option, leave it disabled so that the external SigmaRL controller can take control.

Run the Controller in Separate Terminals

You will usually use three terminals. Every terminal is independent, so environment variables and source commands must be run again in each terminal.

Terminal 1: Optional Zenoh Router

This terminal is only needed if your setup uses Zenoh.

source /opt/ros/jazzy/setup.bash
ros2 run rmw_zenoh_cpp rmw_zenohd

Terminal 2: RViz

Open a new terminal and run:

export WORKSPACE_DIR="$HOME/Documents/ros2-workspace-lab"
cd "$WORKSPACE_DIR"
export ROS_DOMAIN_ID=21
source /opt/ros/jazzy/setup.bash
source "$WORKSPACE_DIR/install/setup.bash"
rviz2 -d ./src/cpm_lab/rviz/default.rviz

Terminal 3: SigmaRL Controller

Open another new terminal and run:

export WORKSPACE_DIR="$HOME/Documents/ros2-workspace-lab"
export SIGMARL_VENV="$HOME/venvs/sigmarl-py312"
export ROS_DOMAIN_ID=1
cd "$WORKSPACE_DIR"
source "$SIGMARL_VENV/bin/activate"
source /opt/ros/jazzy/setup.bash
source "$WORKSPACE_DIR/install/setup.bash"
ros2 launch cpm_lab_sigmarl_control sigmarl_control_planner.launch.py vehicle_ids:=[1,2,3,4] period:=100 model_path:=<path-to-policy-output>

The vehicle_ids argument must match the vehicles started in the simulation. The period argument should match the bridge period configured in the Control Center. The model_path argument should point to the trained SigmaRL policy folder that you want to evaluate in the CPM Lab workflow.