ROS 2 diagnostics tool catches vision-guided robots in the act

A new open-source ROS 2 package called **ros2_kinematic_guard** is exposing why vision-guided robots still fail in production—even when their perception systems technically work. Developed by an engineer identified as zc_Liu, the tool addresses a critical gap in robotics diagnostics: systems often report healthy status (high confidence, valid TF trees, normal topic publishing) while their physical execution paths silently diverge from expected paths due to unmonitored factors like thermal drift, timing inconsistencies, or action instability.

The package introduces three lightweight residual monitoring methods: **Pose Residual Drift Monitoring** tracks divergence between expected and executed poses over time, crucial for detecting mechanical/thermal calibration drift; **Temporal Coherence Residual Tracking** checks timestamp alignment across image frames, TF transforms, and grasp pose generation to catch async ordering issues; and **Action Stability Residual Detection** identifies oscillations or jitter in generated grasp points or servo actions that might indicate instability under lighting or reflection disturbances. This shifts the diagnostic focus from 'Did perception succeed?' to 'Did the system remain converged throughout the entire perception → planning → action pipeline?'