The automation of warehouse systems for a precise flow of goods in distribution centers is still a big challenge and up to now the order picking process is mainly done by a large extent of manual work.
In order to robotize this process, many complex processes such as computation of mixed pallets, vision recognition and material handling must be handled.
A new concept called AUTOPICK for automated order picking with industrial robots is now presented. The case of a gross distribution centre for groceries and food products is used as an example to explain the full process and the proposed solution.
 |
 |
| Figure 1: Autopick robot on shuttle transfer | Figure 2: example of the real pallet |
|
The main task of the "AUTOPICK" consists of building mixed pallets by an industrial robot, mounted on a shuttle transfer. Figure 1 shows the idea of a prototype cell for the feasibility study and Figure 3 shows the developed robot platform. The row of pallets can be two as in the example, or just one. The shuttle moves in a pick aisle from pick slot to pick slot (or from pallet to pallet) and the robot selects the desired items to palletize them on the shuttle pallet. The picking structure can be composed by two simple pallets on ground with stop positions, or by two mobile carousels able to supply a higher number of selectable items / articles. In case the system is at a high handling, the two lines can be formed by a gravity warehouse on a few levels that can receive supplies from its rear side. |
 Figure 3: robot palletizing |
|
Considering the system of figure 1, in order to fulfill this task the following hardware components are required:
The current implementation consist of two different phases. In the first phase, the offline preparation phase, a purchase order can be specified by an operator or by the mainframe system assisted by operator. Based on the available information in the warehouse management system and the calculated pallet, a picking robot program is generated in the final step. This program can then be downloaded onto the robot controller. |
Figure 4: 3D Layout
|
In the second phase, the online picking phase, the robot program can be executed. According to the number of items to be palletized the following steps take place:
- Shuttle movement to the desired "pick position" along the 7° axis slide.
- Robot movement inside the pick area.
- Vision based recognition of position and orientation of the items in the current top layer.
- Plausibility check for all locations and trajectory computation:
- Measured coordinates must be in allowed and expected Z-distance and orientation ranges.
- Check for potential collisions between the robot / gripper and control cabinet and racks.
- Check for potential work space violation of the robot during picking.
- Trajectory computation.
- Adaptation of pick orientation depending on the pick position.
- Picking of item.
- Robot movement out of pick slot. If the robot has reached a certain point over the shuttle pallet, the shuttle is simultaneously sent to the next pick slot, if a different item has to be selected.
- Placement of item on shuttle pallet.
 |
 |
| Figure 5: Layout of the feasibility work cell with one pick aisle and 10 pick slots (lengths are in inches). | Figure 6: Overview of software |
Performance
In order to evaluate the performance, several test runs have been made. One example is shown in Figure 7.
The overall performance can objectively be characterized by the averaged cycle time for picking and placing of one item.
Here is an example of the range of possible times:
|
 Figure 7: Screen of virtual mixed pallet |
The recognition performance depends on the number and quality of the features. Unfortunately, a lot of products are packed into blank cartons without any features on the top side. Cartons with good (black) and many features (printings) would help a lot to improve the stability and performance of the depalletizing system.
 |
|
| Figure 8: Robot & Slide scheme |
In this paper, we have introduced a new approach for automated order picking with industrial robots based on the Autopick principle. Several experiments demonstrate the possibility to automate this process.