Deteksi Visual Berbasis AI pada Sistem Persepsi Robot Humanoid

Abstract

Visual perception is one of the fundamental capabilities required for humanoid robots to interact intelligently with
their environment. This study presents the implementation of AI-based vision detection using convolutional neural
networks (CNN) and real-time inference to enhance the object recognition capability of humanoid robot systems.
The objective of this research is to evaluate the performance of AI vision models in detecting objects under various
lighting conditions and viewing angles. The methodology includes dataset acquisition, preprocessing, model
training, and embedded inference testing on a humanoid prototype. Experimental results demonstrate that the
integrated AI vision model improves accuracy and detection response time, enabling the humanoid robot to
perform navigation and object interaction tasks more effectively. This work contributes new insights into the
integration of modern AI vision techniques in humanoid perception systems using lightweight neural architectures
suitable for real-time robotics.