Sensor Fusion
Combining IMU, GPS, cameras, LiDAR into one estimate.
Why it matters in robotics
Almost every real robot fuses noisy, asynchronous sensors (IMU, GPS, wheel odometry, cameras, LiDAR) into one consistent state estimate, so interviewers use sensor fusion to probe whether you truly understand probabilistic state estimation rather than just calling a library. Expect to derive or reason about the Kalman/EKF predict-update cycle, explain how covariances weight each sensor, and discuss practical issues like time synchronization, frame conventions, and outlier rejection. Strong answers connect the math to a concrete stack (e.g. robot_localization or a VIO pipeline) and to failure modes like GPS dropout or IMU drift.
Application focus
The same topic, tailored to the robot you're building. Your choice is remembered across the roadmap and every topic.
At a glance
The recursive Bayes-filter loop at the heart of sensor fusion: a motion model predicts the next state, each sensor measurement corrects it, and the fused estimate feeds back as the next prior.
What to study
- ✓Bayes filter foundations and the predict-update cycle: KF, EKF, and UKF, plus when linearization breaks down
- ✓How measurement and process noise covariances weight each sensor, and why consistent, well-tuned covariances matter as much as the mean
- ✓Loosely vs tightly coupled fusion and filtering vs smoothing/factor-graph approaches (e.g. EKF-based VIO vs optimization-based VINS)
- ✓Practical engineering: time synchronization, coordinate frames and extrinsic calibration, IMU bias/drift, GPS dropout, and outlier/chi-squared gating
Study by time budget
Pick the path that fits the time you have before your interview.