Real-Time Industrial Bin-Picking with a Hybrid Deep Learning-Engineering Approach

The real-time pick and place of 3D industrial parts randomly filed in a part-bin plays an important role for manufacturing automation. Approaches based solely on the conventional engineering discipline have been shown limitations in terms of handling multiple parts of arbitrary 3D geometries in real-time. In this paper, we present a hybrid approach of deep learning and engineering as a means of breaking through the current limitations of industrial bin picking toward enabling the real-time pick and place of multiple 3D parts of arbitrary geometries. The proposed approach, first, makes use of deep-learning based object detectors configured in a cascaded form for both detecting parts in a bin and extracting features associated with the individual parts detected. The concatenation of the part label and the feature labels and their positions associated with the part allows the subsequent part net to have its part recognition rate close to 100%, Furthermore, the part features and their positions are to be fed directly into the estimation of the 3D pose of the corresponding part in a bin as well as its degree of occlusion. The experimental results demonstrate that the proposed approach is able to perform a real-time multiple part bin picking operation for multiple 3D parts of arbitrary geometries with a high precision.