E! 10146, SOMEWAIR, RFND Technologies AB

Important results from the project

A prototype robot collector which picks battery-containing items from waste electronics (WEEE) was built and then tested in realistic conditions. A standard RGB-D camera and suitable lights were used to film objects on the conveyor. The images were segmented, items of interest were classified, and picked with a robot. Research focus was on minimizing the classifier training phase and optimizing classification despite lighting variation and clutter.

Expected long term effects

Results have been overall positive. The planned tasks have all been performed and results as hoped for. There is a solution foundation with all the planned characteristics, a success for the project. As planned a number of improvement areas have been identified. Classifier training time can be further reduced by using a depth camera for detecting object frames. Also, a background simulation tool could be added to make the classification more robust. The hardware can possibly be manufactured more cost-efficiently and made easier for transportation.

Approach and implementation

The project has been carried out in close cooperation with potential end-customers and with a lot of communication between the three project companies. Visits at the Stena sorting sites in Vissenbjerg and in Halmstad, customer interviews, project meetings both at Refind and at DTI, along with frequent phone- and e-mail conversations. Two demo units for image collection were built (for Refind and for DTI), to optimise development and tests. Robot-tests were done in Denmark, for practical reasons.