Towards automatic visual fault detection in highly expressive human-like animatronic faces with soft skin

Abstract

Designing reliable, humanoid social robots with highly expressive human-like faces is a challenging problem. Their construction requires complex mechanical assemblies, large numbers of actuators and involves soft material. When deploying these robots in the field they face problems of wear and tear and mechanical abuse. Mechanical defects of such faces can be hard to analyze automatically or by manual visual inspection. We propose a method of automatic visual calibration and actuator fault detection for complex animatronic faces. We use our approach to scan three expressive animatronic faces, and analyze the data. Our findings indicate that our approach is able to detect faulty actuators even when they contribute to the overall expression of the face only marginally, and are hard to spot visually.