class="info-alert">『Your web browser does not support JavaScript, but it does not affect browsing through the rest of the web site.』
jump to main content

Industrial Technology Research Institute

:::

Robot self-learning technology

Technology Overview

Through the robot simulator, the self-learning software for robot arms developed by ITRI could spend just one night around 12 hours to adopt the new workpiece on the machine.
Through the robot simulator, the self-learning software for robot arms developed by ITRI could spend just one night around 12 hours to adopt the new workpiece on the machine.

Based on deep reinforcement learning technology, a self-learning robot can learn the picking and placing skills for workpieces by itself. In the real world environment, our technology could achieve more than 90% success rate for robot picking randomly placed rigid workpieces.

Applications & Benefits

The metal processing industry is one of the major industries for industrial robots in Taiwan. Since most of the businesses in this industry are small and medium-sized enterprises, the degree of automation is low. Furthermore, the shortage of labor is becoming more serious and the demand for customized products is increasing. Above trends lead to that the automation and flexibility of production lines become the critical goals to be pursued. To help the industry to improve production efficiency, CITC in ITRI develops "robot self-learning technology" to fulfill industrial needs, and it combines AI technology to enable robotic arms to learn the picking and placing for workpieces quickly and flexibly. This technology could cater the demand for small quantity but high variety production.

At present, the traditional workpiece picking and placing modes roughly could be divided into two categories: for visual analysis technology and special purpose machine. If the method for visual analysis technology is adopted, engineers with image processing expertise are required to adjust the computer parameters according to the algorithms. Depending on the complexity of the workpiece type, it could take at least one day or up to seven days to complete the workpiece picking and placing successfully. For the workpiece with mass demand, customized machines can be used for complete picking and placing tasks. However, in the real industrial application, the production is small quantity but high variety and the workpieces are more diversified, so the setting of special machines or the adjustment of visual parameters could lead to high cost and time consuming. Furthermore, the low flexibility issue for special purpose machines might keep the machines in idle. Through the robot simulator, the self-learning software for robot arms developed by CITC in ITRI could spend just one night around 12 hours to adopt the new workpiece on the machine, and then successfully apply the proper parameters to the arm with small adjustment for production, which greatly reduces the lead time to change the workpieces.



Related Links
[{"text":"internship","weight":13.0},{"text":"Collagen Corneal Shield","weight":11.5},{"text":"2019 International Symposia on VLSI","weight":9.4},{"text":"5G","weight":8.0},{"text":"CES","weight":5.0},{"text":"CNC CONTROLER","weight":5.0},{"text":"annual report","weight":5.0},{"text":"cosmetics","weight":5.0},{"text":"drone","weight":4.0},{"text":"EOSL","weight":4.0},{"text":"Map","weight":3.0},{"text":"librarian","weight":3.0},{"text":"metrology","weight":3.0},{"text":"explosion","weight":3.0},{"text":"iKNOBEADS","weight":3.0},{"text":"2.5D","weight":2.0},{"text":"pecola","weight":2.0},{"text":"pesticides","weight":2.0},{"text":"rubber","weight":2.0},{"text":"smart grid","weight":2.0},{"text":"solar","weight":2.0},{"text":"textile","weight":2.0},{"text":"Roll-to-Roll","weight":2.0},{"text":"Smart R","weight":2.0},{"text":"2019 Janssen Taiwan Grant Call","weight":2.0},{"text":"2019 project","weight":2.0}]