Content still under development!
Try at your own risk!
- Repo link: airou-lab/arc_rl_interface
- For Methodology
Passive Camera only RL navigation
Train and evaluate a recurrent policy (RecurrentPPO with CNN+LSTM) that learns autonomous maneuvers from RGB frames only streaming from a Unity mini‑city. No semantic masks, optical flow, handcrafted path planners, or action blending — purely passive visual navigation.
Run model to real world bridge.
We’ve provided a simple script to run a trained model on the robot. Simply run the /arcpro_rl.sh in the project’s home directory like so
./arcpro_rl.shInside you will find it calls upon a certain script
Run the unity sim model
Running the model is a bit more complex. Currently, we use
- Unity version 6000.3.1f1
TODO CONTINUE MODEL RUN SETUP FROM SCRATCH!
Our submodel is linked as so: arc_rl_interface