119 lines
4.4 KiB
Markdown
119 lines
4.4 KiB
Markdown
|
---
|
||
|
|
title: "MegaTag — Robot Localization with AprilTags"
|
||
|
|
tags:
|
||
|
|
- megatag
|
||
|
|
- apriltag
|
||
|
|
- vision
|
||
|
|
- localization
|
||
|
|
- limelight
|
||
|
|
- photonvision
|
||
|
|
- odometry
|
||
|
|
- frc
|
||
|
|
- 2890
|
||
|
|
type: training-module
|
||
|
|
owner: 2890
|
||
|
|
status: active
|
||
|
|
sources:
|
||
|
|
- "https://docs.limelightvision.io/docs/docs-limelight/pipeline-apriltag/apriltag-robot-localization"
|
||
|
|
- "https://docs.yagsl.com/fundamentals/swerve-drive"
|
||
|
|
---
|
||
|
|
|
||
|
|
# MegaTag — Robot Localization with AprilTags
|
||
|
|
|
||
|
|
## What Is MegaTag?
|
||
|
|
|
||
|
|
MegaTag is a **Limelight-specific** robot localization system that uses AprilTags + IMU (gyro) data to determine the robot's position on the field. It's not a Limelight-only concept — similar approaches exist with PhotonVision + WPILib pose estimators — but "MegaTag" specifically refers to Limelight's implementation.
|
||
|
|
|
||
|
|
**Key point:** Team 2890 uses **PhotonVision**, not Limelight. MegaTag documentation is useful for understanding the **concept** of vision-based odometry fusion, but the implementation uses PhotonVision's pose estimator + WPILib's `SwerveDrivePoseEstimator`.
|
||
|
|
|
||
|
|
## MegaTag vs. MegaTag2
|
||
|
|
|
||
|
|
| Feature | MegaTag 1 | MegaTag 2 |
|
||
|
|
|---------|-----------|-----------|
|
||
|
|
| Data source | AprilTag vision only | AprilTag + IMU fusion |
|
||
|
|
| Accuracy | Good when tags visible | Better — gyro corrects drift |
|
||
|
|
| Drift | Accumulates over time | Reduced by gyro correction |
|
||
|
|
| Tag visibility required | Yes | Yes (but less sensitive) |
|
||
|
|
|
||
|
|
**MegaTag 2** fuses robot orientation data from the IMU with vision pose. This means even if vision is slightly noisy or a tag is partially visible, the gyro steadies the reading.
|
||
|
|
|
||
|
|
## How It Works (Conceptual)
|
||
|
|
|
||
|
|
```
|
||
|
|
AprilTag camera → Robot pose (x, y, rotation)
|
||
|
|
↓
|
||
|
|
Gyro heading → Corrections / steadying
|
||
|
|
↓
|
||
|
|
Pose Estimator → Combines:
|
||
|
|
- Wheel odometry (always drifting)
|
||
|
|
- Vision pose (accurate but intermittent)
|
||
|
|
↓
|
||
|
|
Final robot position (fused)
|
||
|
|
```
|
||
|
|
|
||
|
|
## For Team 2890 (PhotonVision)
|
||
|
|
|
||
|
|
Team 2890 runs PhotonVision on Raspberry Pi, AprilTags on the field. The equivalent fusion happens in WPILib:
|
||
|
|
|
||
|
|
```java
|
||
|
|
// SwerveDrivePoseEstimator fuses:
|
||
|
|
// 1. Wheel odometry (always running, drifts)
|
||
|
|
// 2. Vision measurements (accurate when tags visible)
|
||
|
|
|
||
|
|
SwerveDrivePoseEstimator estimator = new SwerveDrivePoseEstimator(
|
||
|
|
kinematics,
|
||
|
|
gyro.getRotation2d(), // IMU heading
|
||
|
|
modulePositions, // wheel encoders
|
||
|
|
new Pose2d(x, y, rotation) // initial estimate
|
||
|
|
);
|
||
|
|
|
||
|
|
// Add vision measurement when tag visible:
|
||
|
|
estimator.addVisionMeasurement(
|
||
|
|
photonPoseEstimator.getEstimatedPosition(),
|
||
|
|
Timer.getFPGATimestamp()
|
||
|
|
);
|
||
|
|
```
|
||
|
|
|
||
|
|
## The Key Concept for Students
|
||
|
|
|
||
|
|
**Robot localization = knowing where you are on the field.**
|
||
|
|
|
||
|
|
AprilTags give you a reference point. The gyro gives you orientation. Fusing them gives you accurate, stable position tracking even when:
|
||
|
|
- Only one tag is visible
|
||
|
|
- Tags are partially obscured
|
||
|
|
- Robot is moving fast
|
||
|
|
|
||
|
|
This is critical for:
|
||
|
|
- Autonomous paths that need to return to same spot
|
||
|
|
- Field-relative driving with joystick
|
||
|
|
- Score estimation / match strategy
|
||
|
|
|
||
|
|
## Common Issues and Troubleshooting
|
||
|
|
|
||
|
|
| Problem | Likely Cause | Fix |
|
||
|
|
|---------|--------------|-----|
|
||
|
|
| Pose always offset in same direction | Camera calibration wrong | Check camera tilt, height, direction |
|
||
|
|
| Pose jumps when tag visible | Vision trusting single reading | Add filtering, check timing |
|
||
|
|
| No pose when tags visible | Camera not processing tags | Check PhotonVision pipeline |
|
||
|
|
| Odometry drifts despite vision | Vision not being added to estimator | Check `addVisionMeasurement` call |
|
||
|
|
| Gyro disagrees with vision | Gyro calibration issue | Re-calibrate gyro, check wiring |
|
||
|
|
|
||
|
|
## Connection to Training
|
||
|
|
|
||
|
|
For students learning swerve odometry:
|
||
|
|
1. **Wheel odometry** — always tracking, always drifting
|
||
|
|
2. **Vision pose** — accurate when AprilTags visible
|
||
|
|
3. **Pose estimator** — fuses both for best of both worlds
|
||
|
|
4. **Field constants** — AprilTag positions, field dimensions must be correct
|
||
|
|
|
||
|
|
The MegaTag concept (vision + gyro fusion) is the same regardless of whether you use Limelight or PhotonVision. Understanding it helps students debug pose estimation issues.
|
||
|
|
|
||
|
|
## Related
|
||
|
|
|
||
|
|
- [[photonvision]] — Team 2890's vision system
|
||
|
|
- [[swere-modules]] — MK4i with encoders for odometry
|
||
|
|
- [[systemcore]] — upcoming controller with improved processing
|
||
|
|
|
||
|
|
---
|
||
|
|
*Research from web search — MegaTag concept for FRC vision localization*
|
||
|
|
*Queue: research complete — stored in wiki*
|