Objective
Independently develop a complete pipeline for a vision-based energy-aware frontier exploration system for a UAV, using open-source components.
Approach
Perception & Mapping -- Integrated ORB SLAM3 with DJI Tello and ROS, to implement monocular SLAM.
​
Navigation -- Due to the absence of an open-source navigation stack for the DJI Tello, I attempted to develop a basic navigation script. However, challenges such as the drone's noisy odometry and limitations of ORB-SLAM3 hindered its functionality, leading me to transition to simulations.
​
Energy-aware Planning -- Modified the explore_lite and move_base ROS packages to implemented an adaptive threshold based approach to maximize exploration while minimizing travel distance. This approach revolves around a cost function with four parameters.
Cost Function: F = W1·Dr + W2·Db + W3·De − W4·Gi
where,
W1, W2, W3, and W4 are weights
Dr is the distance of the frontier to the robot
Db is the distance of the frontier to the base station
Gi is the size of the frontier
De is simply a parameter that prioritizes further frontiers when the battery is higher and when the battery is lower it prioritizes frontiers closer to base
​