Implement ros2_svdd_monitor: SVDD anomaly detection for proprioceptive sensors

[Copilot]

Implements a complete ROS 2 Python package for real-time anomaly detection using Support Vector Data Description (SVDD via OneClassSVM) on proprioceptive sensors (/cmd_vel and /imu). Detects anomalies by learning the expected relationship between commanded motion and IMU response.

Package Structure

• Core modules: SVDD model wrapper (svdd_model.py), sliding-window feature extraction (features.py), ROS 2 monitor node (monitor.py), training CLI (train.py)
• Utils: rosbag-to-CSV converter for data export
• Config: YAML-based hyperparameters (window_size, nu, gamma, thresholds)
• Tests: 17 unit tests covering feature extraction and model operations

Entry Points

train: Offline training from CSV

ros2 run ros2_svdd_monitor train --csv normal_operation.csv
# Extracts 20 features (cmd_vel stats + IMU stats + correlations)
# Trains OneClassSVM, saves model + scaler

monitor: Real-time anomaly detection node

ros2 run ros2_svdd_monitor monitor
# Subscribes: /cmd_vel, /imu
# Publishes: /svdd/anomaly (Bool), /svdd/anomaly_score (Float32)

Feature Engineering

Extracts 20 features from sliding windows:

• 8 cmd_vel features: mean/std/min/max of linear.x and angular.z
• 8 IMU features: mean/std/min/max of accel.x and gyro.z
• 4 cross-features: correlations and ratios mapping commands to expected sensor response

Detects wheel slip, motor failures, collisions, sensor degradation.

Example Usage

# Generate training data
python examples/generate_sample_data.py --samples 100

# Train model
ros2 run ros2_svdd_monitor train --csv example_training_data.csv

# Monitor in real-time
ros2 run ros2_svdd_monitor monitor

# View results
ros2 topic echo /svdd/anomaly
ros2 topic echo /svdd/anomaly_score

Changes

• Added complete ament_python package structure under src/ros2_svdd_monitor/
• Implemented SVDD model wrapper using sklearn.svm.OneClassSVM with RBF kernel
• Created feature extraction with configurable sliding windows and cross-feature correlations
• Built ROS 2 node for real-time monitoring with anomaly score publishing
• Added training CLI with CSV input and joblib model persistence
• Configured pytest CI workflow with secure permissions
• Updated LICENSE from Apache 2.0 to MIT
• Preserved .devcontainer/ from template
• Added examples with synthetic data generator

Configuration

All hyperparameters in config/config.yaml:

window_size: 10
nu: 0.1  # outlier fraction upper bound
gamma: 'scale'  # RBF kernel coefficient
anomaly_threshold: 0.0

Original prompt

I want a ROS 2 (Humble) Python package seeded from the LCAS/ros2_pkg_template. Create a repository and package named ros2_svdd_monitor (owner: LCAS) that implements an SVDD-style anomaly detection pipeline using ONLY proprioceptive sensors: /imu and /cmd_vel. Preserve the .devcontainer from the template. Target rclpy (Python) with console entry points train and monitor.

Requirements:

Create a ROS 2 ament_python package ros2_svdd_monitor under src/ with setup.cfg/setup.py/package.xml and entry points:
train: CLI that reads a CSV (exported from rosbag), extracts sliding-window features from /cmd_vel + /imu, scales features, trains sklearn.svm.OneClassSVM (RBF) as the SVDD proxy, and saves model+scaler (joblib).
monitor: rclpy node subscribing to /cmd_vel and /imu, maintaining configurable sliding windows, extracting features, loading the saved model, publishing /svdd/anomaly (std_msgs/Bool) and /svdd/anomaly_score (std_msgs/Float32).
Implement svdd_model.py (OneClassSVM wrapper), features.py (feature extraction and mapping logic: map cmd_vel.linear.x forward command to expected IMU forward accel/orientation response), utils/rosbag_to_csv.py (helper guidance or minimal script to export topics), config/config.yaml (window_size, nu, gamma, model_path, threshold).
Tests: unit tests for feature extraction and model save/load (pytest).
CI: GitHub Actions workflow to run pytest (and lint optionally) on push/PR.
requirements.txt listing scikit-learn, numpy, pandas, joblib, pyyaml, rclpy.
License: MIT.
Do NOT include rosbag files in the repo. Provide example usage in README and a short note on how to record/export CSV from rosbag (recommend a small subscriber to write synchronized CSV rows).
Commit all files to main with message "Initial ros2_svdd_monitor package (SVDD monitor)". Make hyperparameters configurable in config.yaml and keep outputs human-readable.
Deliverables: complete package files, tests, CI, README with quickstart for training and running the monitor, and devcontainer preserved for local development.

---

💡 You can make Copilot smarter by setting up custom instructions, customizing its development environment and configuring Model Context Protocol (MCP) servers. Learn more Copilot coding agent tips in the docs.