Physics-Informed Residual Learning for Adaptive Friction Compensation

A research project developing Sim-to-Real adaptive control frameworks using Physics-Informed Neural Networks (PINNs) to compensate for nonlinear friction dynamics in low-cost actuators.

📋 Project Overview

This research addresses the challenge of precise motion control in cost-constrained robotic systems where traditional PID controllers fail to capture complex friction phenomena (Stribeck effect, stiction, Coulomb friction). The core innovation is learning the residual dynamics — the "rust" between ideal models and real hardware — rather than full system identification.

Key Contributions

• Physics-Informed Learning: PINN architecture constrained by Stribeck and LuGre friction models
• Sim-to-Real Transfer: Pre-training in differentiable physics simulators before hardware deployment
• Edge Deployment: Sub-10ms inference latency on embedded targets (Teensy 4.1)
• Hybrid Control: Combines classical PID with learned residual compensation

🎯 Technical Constraints

Constraint	Requirement	Rationale
Latency	< 10ms control loop	Real-time performance for stable control
Compute	Edge-only inference	No PC-in-the-loop; standalone operation
Dynamics	Single-joint focus	Proof-of-concept before multi-DOF extension
Learning	Sim-to-Real pre-training	Safety; no unsafe exploration on hardware

📂 Repository Structure

PIRL_Adaptive_Control/
├── sim/                          # Differentiable physics simulators & training
│   ├── controllers.py           # Baseline controllers (PID, PPO, etc.)
│   ├── pirl_network.py          # Physics-Informed Neural Network
│   ├── sim_env.py               # Gym-like simulation environment
│   └── train_*                  # Training scripts
├── firmware/                     # Embedded C++ code for Teensy 4.1
│   ├── src/main.cpp             # Main control loop
│   ├── lib/                     # Hardware abstractions
│   └── test/                    # Unit tests and validation
├── notebooks/                    # Exploratory analysis & visualization
│   ├── 01_Deriving_Stribeck.ipynb
│   ├── 02_Differentiable_Physics.ipynb
│   └── 03_Stribeck_PINN.ipynb
├── models/                       # Trained model checkpoints
├── scripts/                      # Export and deployment scripts
├── docs/                         # Theory derivations and manuscripts
└── data/                         # Training and validation datasets

🛠️ Installation

Quick Start (Simulation Only)

# Clone repository
git clone https://github.com/basyirin-dev/PIRL_Adaptive_Control.git
cd PIRL_Adaptive_Control

# Install Python dependencies
pip install -r requirements.txt

# Run basic simulation test
python -m sim.test_sim

Full Installation (Including Firmware)

# Install PlatformIO for embedded development
pip install platformio

# Build and test firmware
cd firmware
pio run -e native_test
pio test -e native_test

🚀 Usage

Training Physics-Informed Models

# Train baseline PID controller
python scripts/train_and_export.py --controller pid --epochs 100

# Train PIRL hybrid controller
python scripts/train_and_export.py --controller hybrid --epochs 200

# Export to ONNX for deployment
python scripts/export_onnx.py --model models/pirl_model.pth

Running Simulations

# Basic simulation test
python -m sim.test_sim

# Generate convergence figures
python sim/generate_figure_1.py

# Run ablation studies
python sim/run_asymmetric_ablation.py

Firmware Deployment

# Build for Teensy 4.1
pio run -e teensy41

# Upload to hardware
pio run -e teensy41 --target upload

# Monitor serial output
pio device monitor -b 2000000

📊 Results

Simulation Performance

• Stribeck Curve Fitting: RMSE < 0.05 Nm
• Velocity Tracking: < 2% error at 10 Hz
• Convergence: Stable within 500 training episodes

Hardware Performance (Teensy 4.1)

• Control Frequency: 100 Hz (10 ms period)
• Inference Latency: < 100 μs
• Total Loop Time: < 1 ms (including sensor I/O)

📚 Citation

If you use this work in your research, please cite:

@misc{pirl_adaptive_control,
  author = {Basyirin Amsyar bin Basri},
  title = {Physics-Informed Residual Learning for Adaptive Friction Compensation},
  year = {2024},
  publisher = {GitHub},
  journal = {GitHub repository},
  howpublished = {\url{https://github.com/basyirin-dev/PIRL_Adaptive_Control}}
}

See CITATION.cff for additional citation formats.

🤝 Contributing

Contributions are welcome! Please follow these guidelines:

1. Fork the repository
2. Create a feature branch (git checkout -b feature/your-feature)
3. Commit changes (git commit -m 'Add your feature')
4. Push to the branch (git push origin feature/your-feature)
5. Open a Pull Request

Development Setup

# Install development dependencies
pip install -r requirements-dev.txt

# Run tests
pytest sim/test_sim.py -v

# Check code formatting
black --check sim/ firmware/

📅 Project Status

• ✅ Phase 1: Simulation & Theory (Complete)
• 🚧 Phase 2: Hardware Validation (In Progress)
• 📋 Phase 3: Multi-DOF Extension (Planned)

📄 License

This project is licensed under the MIT License - see the LICENSE file for details.

🔗 Links

• GitHub Repository
• Issue Tracker
• Changelog

📧 Contact

For questions or collaboration opportunities, please contact:

• Principal Investigator: Basyirin Amsyar bin Basri
• Repository: https://github.com/basyirin-dev/PIRL_Adaptive_Control

---

Note: This is an active research project. APIs and implementations may change as the project evolves.