SLAM
Simultaneous Localization And Mapping — the crown jewel.
Why it matters in robotics
SLAM is the canonical hard problem in robotics perception and the one interviewers use to test whether you truly understand probabilistic state estimation end-to-end. Expect to be pushed on the predict-correct cycle, why naive EKF-SLAM scales as O(n^2) in landmarks, the role of loop closure and data association, and the modern front-end (feature tracking) versus back-end (graph optimization) split. Strong answers connect the math (Gaussians, factor graphs, marginalization) to real failure modes like drift and perceptual aliasing.
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 core SLAM estimation loop: motion-model prediction, sensor-based correction, and loop closure to correct accumulated drift.
What to study
- ✓Filtering SLAM: EKF-SLAM and the FastSLAM/particle-filter approach, plus why the covariance grows quadratically with landmarks
- ✓Graph-based / factor-graph SLAM: pose graphs, bundle adjustment, marginalization, and nonlinear least-squares back-ends (g2o, GTSAM, Ceres)
- ✓Front-end essentials: feature detection/matching, data association, motion models, and loop-closure detection (place recognition)
- ✓System view: visual vs LiDAR SLAM, drift and loop-closure correction, and the robust-perception challenges from Cadena et al.
Study by time budget
Pick the path that fits the time you have before your interview.