Person Tracking for Robot Behaviors

Source: ros2-copilot-skills person tracking skill

Why This Matters

Detecting a person once is not the same as tracking a person through time. Robots that react to humans need continuity, position estimates, and confidence over time. Otherwise every frame becomes a new argument about whether the robot really saw someone.

Distilled Takeaways

• Tracking is about identity and continuity, not just repeated detection.
• Spatial tracking in map or robot frames is far more useful for robot behavior than raw image coordinates.
• Even simple nearest-neighbor or Kalman-based tracking can be enough if the problem is scoped correctly.
• Track confirmation and coast time are essential to avoiding twitchy behavior.
• Privacy and retention decisions are part of the design when people are involved.

Practical Guidance

• Separate detection, tracking, and decision logic into different layers.
• Publish a tracked-person message that downstream behaviors can consume directly.
• Require a track to survive multiple frames before letting it trigger autonomy.
• Define how long a lost track should persist based on your robot's environment and latency budget.

Corroborating References

• vision_msgs package
• BehaviorTree.CPP docs

When to Read the Original Source

Go to the original skill when you want concrete matching logic, simple track data structures, patrol-oriented BT integration patterns, and the privacy cautions that matter for in-home or human-facing robots.