| Abstrakt: | This thesis presents a robotic system for reproducing simple black-and-white sketches
using the semi-humanoid robot NICO. The proposed system integrates visual percep
tion, graph-based sketch representation, path planning, and motion execution into a
unified pipeline that transforms a captured sketch into a physical drawing on a digital
tablet.
The perception stage focuses on reliable whiteboard detection and geometric recti
fication using a marker-based approach. The extracted sketch is processed through a
graph construction pipeline consisting of binarization, skeletonization, and vectoriza
tion. The resulting graph representation is then used for trajectory generation, where
several path planning strategies are explored to produce coherent and efficient drawing
motions. Finally, the generated trajectories are executed on the humanoid robot using
inverse kinematics and motion control techniques.
We evaluated the system on a custom dataset of hand-drawn sketches captured
under varying distances, orientations, and lighting conditions. Experimental results
demonstrate that the proposed pipeline achieves reasonably reliable sketch reproduc
tion while maintaining near real-time performance. The work shows that a modular
combination of classical image processing, graph-based methods, and robotic motion
control can successfully address the constrained task of robotic sketch reproduction.
|
|---|