Multi-object recognition on the basis of vision and manipulation

The authors aim at recognition of multiple objects by using robot vision and its manipulation. In the case where multiple objects are arranged in a shelf, front objects may conceal some objects behind them. A robot can recognize these objects if it can recognize the front objects using prior knowledge and remove them with its arm. However, it sometimes fails to recognize the objects because of the lack of good features, a bad lighting condition etc. Here, we proposed a new method of the multi-object recognition using the manipulation. Recognized objects are removed using pickup motion. Unrecognized objects are pushed for changing the angle when the robot cannot recognize them. We show the experimental results of our proposed method. From the results, we can confirm that the manipulation is useful for improving the multi-object recognition.