Problem
Task 1 - Autonomous Mapping Enable a mobile robot to autonomously explore and map an unknown environment without fixed waypoints or prebuilt navigation stacks, under a time constraint
Task 2 - Navigation with Static Obstacles Navigate to a user-given goal pose using a prebuilt map, while avoiding static obstacles that are not present during mapping, and will spawn randomly
Task 3 - Perception Driven Search and Localization Detect RGB colored objects while autonomously navigating through the prebuilt map, and publish their estimated positions
Approach
Task 1 - Autonomous Mapping Designed a reactive exploration strategy combining RRT based frontier detection, obstacle avoidance, and stochastic motion to maximize map coverage using SLAM
Task 2 - Navigation with Static Obstacles Implemented a custom A-star global path planner on occupancy grids, integrated with localization (AMCL) and a low-level path following controller, with a local RRT-star path planner for avoiding the obstacles that are not on the occupancy grid
Task 3 - Perception Driven Search and Localization Built a vision based perception pipeline using color segmentation and geometric reasoning to detect objects and estimate their positions in the global frames
Solution
Task 1 - Autonomous Mapping Developed a filly autonomous optimal mapping node, that generated occupancy grid maps with high coverage, saving and loading the maps
Task 2 - Navigation with Static Obstacles Created a collision free navigation system that dynamically adjusted motion in the presence of new obstacles while maintaining progress toward assigned goals
Task 3 - Perception Driven Search and Localization Successfully detected red, blue and green objects and published their positions to dedicated topics, with bounded localization error
Impact
Task 1 - Autonomous Mapping Achieved >90% coverage in under 10 minutes, and stood 2nd in the class for the amount of coverage in a given time
Task 2 - Navigation with Static Obstacles Developed and algorithm that generated a global path in under 0.5 seconds and an RRT-star local path in under 1 second for any given goal pose across the map
Task 3 - Perception Driven Search and Localization Demonstrated a flawless perception-planning-control pipeline that detected all the balls under 6 minutes, and stood first in the class for having the fastest algorithm for the same