Sensors & How They Work
IMUs, LiDAR, cameras, GPS — how robots perceive.
Why it matters in robotics
Sensors are the front end of every robot autonomy stack, so interviewers probe whether you understand what your data physically represents before it ever reaches a filter or planner. Expect questions on why IMUs drift (bias, scale-factor error, random walk), how cameras and LiDAR are calibrated (intrinsics, extrinsics, time sync), and how GPS, encoders, and range sensors fail in the real world (multipath, wheel slip, dropout, specular surfaces). A very common pattern is "your estimate is drifting or jumping, which sensor is to blame and how would you diagnose it?", which tests whether you can reason about noise models and failure modes rather than recite specs. Strong answers connect a sensor's error characteristics to downstream consequences, like why you cannot dead-reckon on an IMU alone or why a single sensor needs complementary fusion. This topic is foundational for state estimation, SLAM, and perception rounds, so weakness here undermines the rest of the interview.
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
How a raw IMU rate measurement turns into unbounded drift, motivating sensor fusion.
What to study
- ✓IMU error model: bias (constant plus bias instability), scale-factor error, white noise vs. random walk (angular/velocity random walk), and why integration turns small rate errors into unbounded position/heading drift.
- ✓Camera and LiDAR operating principles plus calibration: pinhole intrinsics and lens distortion, time-of-flight/structured ranging, and intrinsic, extrinsic, and temporal (time-sync) calibration via checkerboards and tools like Kalibr.
- ✓Other sensors: wheel encoders (resolution, slip, dead-reckoning error growth), GPS/GNSS (multipath, urban-canyon dropout, ~meter-level absolute accuracy, RTK), and range sensors (ultrasonic/ToF range and specular-surface limits).
- ✓Noise models and failure modes: zero-mean Gaussian measurement noise, reading a datasheet and Allan deviation plot, and the characteristic failure mode of each sensor that motivates multi-sensor fusion.
Study by time budget
Pick the path that fits the time you have before your interview.