https://github.com/RTIS-Lab/rtis_f110_rl
From the C++ trainer, there’s a python script at tools/export_traced.py. Run it with a trained model as: python tools/export_traced checkpoints/run_name/run_STEPS. Replace STEPS with the step count or final if it finished training. Do not include the _actor.pt suffix - the script will do it for you. It will generate a run_STEPS_traced.pt file.
Copy this to the package’s models folder, and set the model_path file in your config file to match.
Ensure fields like model_speed_cap, model_speed_min, num_frames, and update_ms are set properly.
The rl_inference container hosts the rtis_f110_rl node, and uses Nvidia’s container runtime and dusty-nv’s image to use CUDA in the container. To use a different configuation file, edit CMD line of the inference container’s Dockerfile. You may need to rebuild the container for this to take effect.
If you’re changing values within a configuration file, it’s just enough to restart the container.