Mobile Robot Navigation and Localization

Project Overview

This project builds a mobile robot navigation system for indoor environments where lighting and geometry can vary significantly.

Responsibilities

• Integrated SLAM, localization, and planning modules into a cohesive ROS2 stack
• Defined map lifecycle from online mapping to map reuse
• Tuned local planner behavior for smoother obstacle avoidance
• Added runtime health checks and recovery behaviors

System Design

Perception

• LiDAR-based scan matching for primary map consistency
• Visual odometry as supporting signal in feature-rich zones
• Sensor fusion at the state-estimation layer for improved robustness

Planning and Control

• Global planner for long-distance route generation
• Local planner with dynamic obstacle handling
• Velocity command shaping for stable cornering and docking

Runtime Robustness

• Recovery behaviors for localization loss and blocked paths
• Automatic planner reset when stale map states are detected
• Logging and replay pipeline for post-run analysis

Challenges and Solutions

Challenge: Drift increased in repetitive corridors

Solution:

• Added loop-closure-friendly waypoints in map workflow
• Improved sensor fusion weighting based on confidence metrics
• Tuned scan matching thresholds for corridor geometry

Challenge: Oscillation near narrow passages

Solution:

• Adjusted local planner cost maps and inflation parameters
• Added direction consistency penalty in short-horizon planning
• Reduced aggressive velocity spikes at turning points

Results

• Higher success rate in long-horizon autonomous navigation runs
• More stable docking and pass-through behavior in constrained spaces
• Better reproducibility across different environment lighting conditions

Next Iteration

• Add semantic constraints from vision-based scene understanding
• Support multi-floor map switching with persistent localization context
• Introduce adaptive parameter sets by environment profile