Create ROS2 SVDD anomaly detection package with OneClassSVM

[Copilot]

Implements SVDD-based anomaly detection for robot monitoring using /imu and /cmd_vel topics. Package structure follows ROS2 Humble ament_python conventions with MIT license.

Package Structure

• config.py: SVDD hyperparameters and topic configuration
• features.py: Windowed statistical features (mean/std/min/max) from sensor streams
• svdd_model.py: OneClassSVM wrapper with StandardScaler and persistence
• train_node.py: Collects normal operation data for configurable duration (default 60s)
• monitor_node.py: Real-time anomaly detection publishing to /anomaly_detected and /anomaly_score

Entry Points

ros2 run ros2_svdd_monitor train    # Train on normal data
ros2 run ros2_svdd_monitor monitor  # Detect anomalies

Feature Extraction

48-dimensional feature vector per window:

• IMU: 6 dimensions (linear_accel xyz, angular_vel xyz) × 4 statistics
• cmd_vel: 6 dimensions (linear xyz, angular xyz) × 4 statistics

Testing

• pytest suite covering config, model training/prediction/persistence, feature extraction
• CI enabled with skip-tests changed to false in ros-ci.yml
• Dependencies (numpy, scikit-learn) added to setup.py install_requires

Configuration

Model persists to ~/svdd_model.pkl (excluded via .gitignore with *.pkl, *.bag, *.db3)

Original prompt

Seed from LCAS/ros2_pkg_template as LCAS/ros2_svdd_monitor. Preserve .devcontainer. Create a ROS2 Humble ament_python package (rclpy) implementing SVDD using ONLY /imu and /cmd_vel. Add train (OneClassSVM) and monitor nodes (entry points: train, monitor), svdd_model, features, config, tests, CI (pytest), requirements, README. MIT license. Do NOT include rosbag files.

---

✨ Let Copilot coding agent set things up for you — coding agent works faster and does higher quality work when set up for your repo.

[cooperj]

2024?

[jondave]

@copilot what is the difference between this PR and PR2 with the branch copilot/create-ros2-svdd-monitor-package-again

[Copilot]

@copilot what is the difference between this PR and PR2 with the branch [copilot/create-ros2-svdd-monitor-package-again](https://github.com/LCAS/ros2_svdd_monitor/tree/copilot/create-ros2-svdd-monitor...

The main differences between this PR and the copilot/create-ros2-svdd-monitor-package-again branch are:

This PR (copilot/create-ros2-svdd-monitor-package):

• Uses config.py module for configuration
• Node files named train_node.py and monitor_node.py
• Simpler structure with 3 main commits focusing on core functionality
• No examples directory or sample data generator

Other branch (copilot/create-ros2-svdd-monitor-package-again):

• Uses YAML-based configuration (config/config.yaml)
• Node files renamed to train.py and monitor.py
• Includes examples/ directory with sample data generator
• Adds utility module (utils/rosbag_to_csv.py)
• Has dedicated pytest workflow (.github/workflows/pytest.yml)
• More comprehensive testing improvements
• Enhanced documentation with examples
• Additional .gitignore entries

The other branch appears to be a more feature-complete iteration with better tooling, examples, and testing infrastructure.