Multi-modal Electronics State Evaluation for Robotic Demanufacturing

State detection is of great importance for evaluating the residual value of end-of(-first)-life electronics, which is often determining for the optimal depth of disassembly. To increase the cost-efficiency and reduce the subjectivity that is inherent to human involvement in the state evaluation process, the presented research developed an automated detection system based on Faster R-CNN-FPN using the fusion of color (RGB) and depth (D) images to identify two most encountered defects in end-of-first-life laptops: missing battery and missing cover. A Cross-Attention Fusion (CAF) module is introduced to enhance the detection accuracy. A dataset containing 513 high-quality RGB and 513 corresponding depth images of laptops has been created and annotated. Experimental results show neck fusion with a CAF module achieves the highest detection mAP of 91.1%, highlighting the potential for automated electronics state detection for vision-guided robotic demanufacturing using data fusion techniques.