Pose inexplicably jumps when auto starts

[clrozeboom]

Describe the bug
When we enable in auto, the pose will inexplicably jump to a random position. It can be off field or on.
I haven't seen it do this any other time than when we start an auto. We have a detection routine in our code where it will stop the auto if it is off field. I don't know if we've seen this behavior at comp, just at home.

To Reproduce
Steps to reproduce the behavior:

1. Start robot code
2. Set the initial pose to a preset
3. Enable auto

• Link to code:
  https://github.com/FRC5010/Reefscape2025/tree/main/TigerShark2025

Expected behavior
The pose will not jump to a random place in auto, but remain where it is set.

Screenshots

Desktop (please complete the following information):

• OS: [e.g. Windows 11]
• Project Information:
• WPILib Information:
  Project Version: 2025.3.1
  VS Code Version: 1.96.2
  WPILib Extension Version: 2025.3.1
  C++ Extension Version: 1.23.2
  Java Extension Version: 1.38.0
  Java Debug Extension Version: 0.58.1
  Java Dependencies Extension Version 0.24.1
  Java Version: 17
  Java Location: C:\Users\Public\wpilib\2025\jdk
  Vendor Libraries:
  ChoreoLib (2025.0.3)
  maplesim (0.3.9)
  PathplannerLib (2025.2.3)
  CTRE-Phoenix (v5) (5.35.1)
  CTRE-Phoenix (v6) (25.2.2)
  photonlib (v2025.1.1)
  ReduxLib (2025.0.1)
  REVLib (2025.0.2)
  Studica (2025.0.1)
  ThriftyLib (2025.0.2)
  WPILib-New-Commands (1.0.0)

Additional context
I have implemented a workaround in the code to handle this situation. When I detect that the pose is off by more than 0.1m (arbitrarily) from the last known pose, I re-run the pose-estimator update after resetting the pose to last known. I've seen this behavior when connected over AP, tethered, and in debug mode.
Our pose estimator is in YAGSL code and we are using a copy of the library instead of the vendordep.
The log output I'm getting doesn't seem to have anything off. Encoders haven't moved on the swerve modules. Yaw is basically 0.

https://github.com/FRC5010/Reefscape2025/blob/main/TigerShark2025/src/main/java/swervelib/SwerveDrive.java:1160
      Rotation2d yaw = getYaw();
// The normal update
      swerveDrivePoseEstimator.update(yaw, getModulePositions());

// Check that the pose hasn't jumped
      Pose2d currentPose = swerveDrivePoseEstimator.getEstimatedPosition();
      if (lastPoseEstimate != null) {
        double distance = Math.abs(lastPoseEstimate.getTranslation().getDistance(currentPose.getTranslation()));

// If we are enabled and a big jump occurs, reset the pose and try again
        if(distance > 0.1 && DriverStation.isEnabled()) {
          DriverStation.reportError("Pose error: Distance " + distance + "  Yaw " + yaw + " Last Pose " + lastPoseEstimate + " to " + currentPose, false);
          Arrays.stream(getModulePositions()).forEach(it -> DriverStation.reportError("Pose error: Module " + it , false));

// Reset the pose and try again
          swerveDrivePoseEstimator.resetPose(lastPoseEstimate);
          currentPose = swerveDrivePoseEstimator.update(yaw, getModulePositions());
          distance = Math.abs(lastPoseEstimate.getTranslation().getDistance(currentPose.getTranslation()));
          if(distance > 0.1) {
            DriverStation.reportError("Pose error: Nope that didn't help", false);
          } else {
            DriverStation.reportError("Pose error: Reset pose to last known", false);
          }
        }
      }

// Store the last known good pose
      lastPoseEstimate = currentPose;

[clrozeboom]

Additional context, we were working on odometry calibration and using only odometry to run autos, so no vision inputs are being given. We tried breakpoints at various places where the pose gets reset and not seeing anything reset the pose unexpectedly. We are using PathPlanner's path-generation on-the-fly to create the path, so we are not seeing it set pose to the start of an auto-path.

[clrozeboom]

Figured this out. We have code from way back that zeros our drive train encoders. This caused the pose estimator to register a change in pose that would be relative to how our robot moved prior to starting auto.