Faults and coordinates

Match the robot’s rejection code to the next check, or look up coordinate and trust conventions.

Rejection lookup

Code	Check next
SCHEMA / MALFORMED	Packet version, shape and bounds
CONFIGURATION / CALIBRATION	Session, field, binding, active intrinsics and mode
TIMESTAMP / STALE / FUTURE	Capture clock domain, timing quality and frame age
DUPLICATE / OLD_EPOCH / RESET	Frame identity and reset boundary
HISTORY / MOUNT_HISTORY	Available robot or moving-mount history at capture time
UNKNOWN_TAG / FIELD_BOUNDS / ROBOT_TILT	Field layout, measured mount and pose geometry
AMBIGUOUS	Whether capture-time heading decisively resolves IPPE alternatives
RESIDUAL / INNOVATION	Image residual or disagreement with the pre-fusion estimate
NO_TAGS	Normal when a fresh valid frame contains no tags

Check the camera page for the frame’s validation/fusion decision. If the connection or clock is failing, use Network and power recovery.

Coordinate quick reference

Quantity	Convention
Robot axes	WPILib NWU: +X forward, +Y left, +Z up
Translation / angles	Metres / right-handed rotation
Pitch	Positive points the forward lens down; negative points it up
Wire quaternion	W, X, Y, Z
Wire pose	Field-to-camera
Field origin	Blue-wall-right, fixed across alliance changes
OpenCV camera axes	+X right, +Y down, +Z forward; converted explicitly by the solver

The robot recovers field-to-robot with:

fieldToRobot = fieldToCamera * inverse(robotToCamera)

Use the Geometry guide for visual mount inspection.

Confidence and trust

Translation uncertainty accounts for distance, geometry, image residual, and motion × timestamp uncertainty, with positive conservative floors. Cameras seeing the same tags close in time are correlated; extra observations are not automatically independent evidence.

Heading remains suppressed. It uses a large finite supported deviation until its separate review gate passes. Revised trust also remains behind a reviewed feature flag.

Capture-time trust and reset rules

Revised trust evaluates observations at their capture time before fusion, degrades with odometry travel, elapsed time and collisions, and uses a consistent reset baseline. Empty measurement lists never sample pose history at NaN. Missing capture-time history is never concealed by substituting the current pose.