IMU — GY-80 9-axis module
Measures acceleration and angular velocity. Depending on its configuration, the GY-80 also includes a magnetometer and a barometer.
Role: Estimate changes in motion and orientation.
Limitation: Bias and measurement noise cause drift when acceleration is integrated over time.
Lab check: Examine accelerometer and gyroscope bias, noise, and calibration.
Wheel encoders — Pololu 42 × 19 mm wheel and encoder set
Measure wheel rotation, allowing wheel displacement and robot motion to be estimated.
Role: Estimate linear movement and turning.
Limitation: Wheel slip, uneven terrain, and encoder quantization introduce errors.
Lab check: Verify counts per revolution, wheel diameter, and wheel spacing.
Depth camera — Intel RealSense D435i
Provides RGB images, stereo depth, and an integrated IMU.
Role: Detect environmental features and estimate motion relative to the surroundings.
Limitation: Depth quality depends on distance, scene texture, lighting conditions, and the environment.
Lab check: Verify the reliable depth range, camera calibration, and timestamp synchronization.
An important detail: The Intel RealSense D435i already contains an IMU. You may therefore have two inertial measurement sources if you also use the GY-80. Their measurements should be calibrated and handled separately unless you deliberately fuse them.
2. How the sensors complement each other
Sensor | Primary contribution | Main error |
|---|---|---|
GY-80 IMU | Fast acceleration and rotation measurements | Bias and drift |
Wheel encoders | Wheel displacement and odometry | Slip and quantization |
RealSense D435i | Visual features and depth-based motion estimates | Depth and visual tracking errors |
The encoders and IMU can provide high-frequency motion estimates, while the camera helps constrain accumulated drift by observing the environment. The exact benefit depends on the localization algorithm and the quality of the measurements.
3. How the EKF combines them
GY-80 IMU
Encoders
D435i
Extended Kalman Filter
Predict motion → compare observations → correct the estimate
Estimated pose
Position (x, y, z) and orientation, as appropriate to the robot and filter
For a ground robot, the EKF might predict motion using wheel odometry and gyroscope measurements, then correct the estimate using visual odometry or other camera-derived observations.
One distinction matters: a raw RGB-D camera does not automatically provide a robot pose measurement. You generally need visual odometry, visual SLAM, or another perception algorithm to extract motion or position information from its images and depth data.
4. Suggested
Test the IMU: Keep the robot stationary and measure bias and noise; then rotate it and compare the angular measurements.
Test the encoders: Move the robot a known distance and compare the measured displacement with the actual distance.
Test the camera: Observe a stationary scene and measure depth at several distances.
Run each localization method separately: Compare IMU-based estimates, wheel odometry, and visual odometry.
Run sensor fusion: Fuse suitable measurements using an EKF and compare the resulting trajectory with a reference trajectory.
The main goal is to demonstrate that the fused estimate is generally more robust than individual estimates, provided the sensors are calibrated, time-synchronized, and supplied to the filter with suitable uncertainty values.