VLA Learn

A collection of runnable learning systems, from small teaching examples to a tested ROS 2/Gazebo autonomous-driving project.

Repository map

Section	Learning method	Purpose
ros2_rl_car/	Reinforcement learning (PPO)	A Gazebo car learns an alternating-curvature track through trial and error. This is the primary self-driving project.
point_reach/	Behavioral cloning and DAgger	A compact 2-D demonstration of imitation learning and covariate shift. It is not the Gazebo car.
learning_examples/	Supervised learning and reinforcement learning	Three focused educational programs with documentation beside their code.
vla-arm-sim/	Independent nested repository	The arm simulator remains at the repository root and keeps its own Git history.

Which car learns to drive?

The accepted Gazebo task is ros2_rl_car. It uses PPO, lidar, speed, heading error, and cross-track error. It does not copy a demonstrator, use behavioral cloning, or train from camera pixels in v1. The RGB camera is published for inspection, while the CPU-friendly policy learns from ray features.

The historical camera behavioral-cloning car was a different project and is not part of the current branch. point_reach/ remains as the repository's small behavioral-cloning/DAgger example.

Setup

The root virtualenv supports the educational examples:

python3.10 -m venv .venv
.venv/bin/pip install -r requirements.txt

The ROS car has additional ROS/Gazebo setup and its own CPU-Torch wrapper:

cd ros2_rl_car
./scripts/setup_venv.sh
./scripts/rl_car doctor
./scripts/rl_car smoke
./scripts/rl_car train --gui --steps 20000 --seed 7

See each section README for its full commands, algorithms, mathematics, and measured results.

Verification

Run the automated regression suite from the repository root:

.venv/bin/python -m pytest -q

Verify the driving integration with the real headless simulator:

cd ros2_rl_car
./scripts/rl_car smoke