NOBLEO TECHNOLOGY · DWARL
Reinforcement learning within the dynamic window
Train a local planner in MuJoCo, measure how it navigates, and bring the same policy to ROS 2 Nav2.
Try the example robot → Use your robot →
Linux x86-64 · ROS 2 Jazzy · Experimental · No pretrained policy included
Start with one command
Install pixi, clone the repository and let setup guide you:
git clone https://github.com/nobleo/dwarl.git
cd dwarl
pixi run setup
Pixi manages Python, ROS and Gazebo. Setup remembers your robot. Every command shows its result and a next step.
Prefer to use your coding agent? Give it the included setup skill with a ready-to-paste prompt.
-
01 · Train
Build robot-sized datasets and train a policy. Start with a tiny pipeline check.
-
02 · Test
Evaluate exact checkpoints and inspect success, collision and per-case behavior.
-
03 · Deploy
Install a complete ONNX bundle and connect the controller to Nav2.
Pick up where you left off
pixi run dwarl status
pixi run dwarl next
DWARL replaces the local controller. Your robot supplies its description, drivers, lidar, odometry and localization. Start in simulation; choose policies using validation evidence rather than training reward alone.