End-to-end autonomous RC car system using vision-based imitation learning on Raspberry Pi 5.

Main project: src/robots/rover/

See src/robots/rover/README.md for complete documentation.

• Project Structure: PROJECT_STRUCTURE.md - Overview of the codebase
• Getting Started: src/robots/rover/README.md - Complete rover documentation

• Training Guide: TRAINING_GUIDE.md - How to train ACT policy
• Deployment Guide: DEPLOYMENT.md - Deploy to Raspberry Pi 5
• Training Output: TRAINING_OUTPUT_GUIDE.md - Understanding training results
• Quick Reference: QUICK_REFERENCE.md - Common commands

• Hardware Options: docs/EDGE_DEPLOYMENT_PLAN.md - Raspberry Pi 5 + Hailo / Jetson Orin
• Quantization Deep Dive: src/robots/rover/docs/quantization/ - Complete quantization workflow

RC Receiver → Arduino UNO → Raspberry Pi 5 → Autonomous Control
    (PWM)        (30Hz)      (Camera + ML)

Hardware:

• RC car with 2-channel receiver
• Arduino UNO R3 (PWM reader)
• Raspberry Pi 5 (data + inference)
• Camera (30fps)

Wiring:

• Brown wire → GND
• Purple wire → Arduino Pin 2 (Steering)
• Black wire → Arduino Pin 3 (Throttle)

1. Validate - Run calibration tool to verify signals
2. Collect - Record driving episodes (camera + PWM)
3. Train - Learn from demonstrations (see below ⭐)
4. Deploy - Autonomous control on-device

cd src/robots/rover
python3 src/record/episode_recorder.py --episode-duration 15 --output-dir ./episodes

# See TRAINING_GUIDE.md for full details
./start_training.sh  # Quick start
# or use tmux for persistent training (see TRAINING_GUIDE.md)

# Quantize and deploy
./deploy_to_pi.sh outputs/lerobot_act/best_model.pth mboels@raspberrypi

# On Raspberry Pi
cd ~/EDTH2025/Erewhon/src/robots/rover
python3 src/inference/act_inference_quantized.py \
    --checkpoint models/best_model_static_quantized.pth \
    --camera_id 0 \
    --arduino_port /dev/ttyUSB0 \
    --control_freq 30

📁 Navigate to src/robots/rover/ for full documentation