Embodied AI Glossary中文

World Frame

世界坐标系Essential

The single, unmoving reference frame shared by an entire scene, to which all other positions can be converted.

The world frame is a reference frame fixed in the environment, independent of the robot's own motion; Modern Robotics calls it the fixed frame or space frame {s}, and it might be anchored to a corner of a room or to a tabletop. A robot system also has a base frame (fixed to the robot's base), a body frame (moving with the robot's body), camera frames, and more; each sensor's readings start out in its own local frame and have to go through a coordinate transformation into a shared reference frame — usually the world frame — before they can be planned against together. Simulators typically have an explicit world root node, such as MuJoCo's worldbody; ROS's REP 105 defines map as the frame fixed to the world, with z pointing up. For an arm bolted to a fixed base, code often just treats the base frame as the world frame directly.

ExampleFor tabletop grasping, the world frame's origin is set on the tabletop with z pointing up: once the camera sees the cup, its coordinates are converted from the camera frame to the world frame using the camera's extrinsics, and the arm plans its grasp from there.

Also called
Space Frame, Fixed Frame
Related
Coordinate Frame · Base Frame · Body Frame · Camera Extrinsics · Coordinate Transformation · REP 105
Sources
Modern Robotics (Lynch & Park, 2017), Ch.3 Rigid-Body Motions
ROS REP 105: Coordinate Frames for Mobile Platforms
MuJoCo Documentation: Modeling

See it in the full glossary →