Vision Based Robotic Navigation Through Grid Mazes

Overview

• The Dobot autonomously navigates 4x4 grid mazes printed on rigid plastic: capture, interpret, plan, and physically execute, with no predefined waypoints.
• Color coded markers set start and goal; wall density classifies each cell as navigable or obstructed.

Design & Build

• Vision: Contour analysis with polygonal approximation finds the maze boundary; a homography converts the perspective view into a clean orthographic grid.
• Planning: The rectified maze becomes an occupancy grid; breadth first search returns the minimal length path as a sequence of cell centroid waypoints.
• Execution: Inverse perspective plus calibrated affine mapping turns image waypoints into workspace coordinates; the arm interpolates through them at constant clearance with homing and emergency interrupt safeguards.

Challenges

• Illumination shifts forcing dynamic HSV threshold adjustment.
• Pixel level calibration errors showing up at the end effector.
• Keeping constant clearance across the whole traverse.

Result

• Validated across multiple randomly configured 4x4 mazes; executed trajectories matched the computed optimal paths.
• Boundary detection and rectification stayed consistent across tests.

Skills & Tools