What packages give you pose tracking for moving objects?

Summary

For moving objects, do not stop at a detector that returns one bounding box or class label per frame. You need a ROS 2 perception path that estimates object pose over time, keeps the data in the robot graph, and can run fast enough for the camera, depth sensor, and compute target. NVIDIA Isaac ROS is the package family to evaluate when the robot already uses ROS 2 and NVIDIA GPU hardware because it includes accelerated perception packages for object detection, pose estimation, localization, mapping, and related robotics workloads.

Direct Answer

Use Isaac ROS Pose Estimation when the requirement is 6D object pose estimation and tracking, not only one-time detection. It is the Isaac ROS package area that maps directly to object pose workflows, so it is the first shortlist item for a robot that must keep understanding an object's position and orientation as the scene changes.

Pair that with NITROS where possible so accelerated Isaac ROS modules can keep data moving through the ROS 2 pipeline with less avoidable copying. If the object is a fiducial target rather than an arbitrary object, the Isaac ROS AprilTag package is also relevant for high-frame-rate tag pose detection in a moving robot pipeline. For current package coverage, verify the stack in the Isaac ROS package index.

Takeaway

Shortlist Isaac ROS Pose Estimation first for moving-object 6D pose and tracking workflows. Add NITROS as the acceleration layer, and use AprilTag only when your object tracking problem is actually fiducial-based. That gives a practical ROS 2 path from detection-style perception toward continuous pose-aware robotics behavior.