Blind Spots Detection in Autonomous Robots using Data Fusion and Deep Learning

The adoption of humanoid robots has surged in recent years, particularly in addressing operational limitations. This study introduces a methodology for detecting blind spots and anomalies in robot motion, using the Nao Robot as a case study.The approach integrates multisensor data with deep learning techniques validated on a proprietary dataset covering diverse real-world scenarios, including dynamic non-visible obstacles. High-frequency sensor sampling (T<inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">0</inf> = 0.2 s) ensures accurate data, while DBSCAN clustering and ANOVA F-value-based feature selection identify critical features like obstacle distances and motion parameters for collision prediction.Deep Learning with Sigmoid and ReLU activations, the ADAM optimizer, and Sparse Categorical Crossentropy loss achieved a 95% F1 score and 99% AUC-ROC, demonstrating robust performance in dynamic environments also in real life environment.