Simultaneous Localization and Mapping (SLAM) is the foundational capability that transforms a remote-controlled machine into an autonomous robot. I specialize in deploying and tuning SLAM systems for competition and research environments.
Mapping with SLAM Toolbox
I use the ROS2 SLAM Toolbox for real-time 2D occupancy grid mapping:
- Online Synchronous SLAM — Real-time map building using scan matching for competition arena mapping
- Loop Closure — Automatic detection and correction of accumulated drift errors as the robot revisits previously mapped areas
- Map Serialization — Saving and loading maps for persistent localization across power cycles
- Dynamic Environment Handling — Strategies for dealing with moving obstacles and changing environments during competition
Localization with AMCL
For localization within known maps, I deploy Adaptive Monte Carlo Localization (AMCL):
- Particle Filter Tuning — Optimizing particle count, distribution, and resampling parameters for accuracy vs. computation trade-offs
- Multi-Sensor Fusion — Combining LiDAR scan matching with wheel odometry for robust pose estimation
- Global Localization — Recovering the robot's position from an unknown initial pose using dense particle distributions
Sensor Fusion
I integrate Extended Kalman Filters (EKF) via robot_localization to fuse
multiple odometry sources:
- Wheel encoder odometry for short-term accuracy
- IMU data for orientation and angular velocity
- Visual odometry from camera systems for drift correction
- Configurable covariance matrices for sensor reliability weighting