In the presence of numerous project participants, university dignitaries, and product developers, the results presentation of the first fully automated counterbalance forklifts took place at Linde Material Handling in Aschaffenburg. Four forklifts were used to demonstrate how this technology could work in the future.
With "real" industrial trucks, it's a bit more complex than with the already widespread, unmanned "underride" robots that communicate with their environment – for example, via the VDA 5050 IT interface. Acceleration, braking, steering and mast control, environmental perception, and safety systems for personnel protection all need to be fully digitized to allow for intelligent overall control.
At Linde MH's development center at its main site in Nilkheim, everything still looks a bit provisional. Two E20-E30 series electric counterbalance forklifts await their deployment, fully loaded with sensors, 3D scanners, HD cameras, and antennas mounted on an aluminum profile on the roof.
CE certification is still pending, explains Stefan Prokosch, initiator of the "Cooperative Autonomous Intralogistics Systems" (KAnIS) project. Six emergency stop switches on the left and right sides of the devices are intended to give the audience a sense of security; the control panel is located on the left side and can be operated from the outside via a screen. A control team with remote controls ensures that no one gets "flattened" here. Especially since nine out of ten of the participating university professors are seated in the first two rows. These professors are significantly involved in KAnIS, a €5 million project funded by the government with €2.8 million, contributing seven student projects, twelve bachelor's theses, six master's theses, and one doctorate.
Hans-Georg Stark, project manager from the Faculty of Engineering at the Aschaffenburg University of Applied Sciences, explains that all the systems must work together. As with other projects, it had become clear that Wi-Fi was no longer sufficient for the communication between the components, whose data is aggregated via an edge computer. A 5G campus network from Ericsson, with a latency of 11 milliseconds, is helping with data transmission. Anyone doubting the real-time accuracy will learn that the noticeable delay appearing on the public monitor is not due to an actual lack of data transfer speed, but rather to the screen display.
During the indoor demonstration of the counterbalance forklifts, the initial task for the premiere audience is simply to move a wire mesh container or pallet from point A to point B. An additional camera on the forks is used, which, via the fork positioner, ensures the correct positioning at the pallet pockets. "With the outdoor tasks," explains Professor Mark Hanke, "we naturally also have to contend with limitations to our perceptual capabilities." Snow, rain, fog, and dirt splashes make it difficult for the instruments to locate and measure distances outdoors.
Professor Klaus Zindler outlines the more complex model routes traversed on the company premises. Model-based position estimates derived from simulations are supplemented indoors – a not entirely new approach – by reflectors, and the actual localization is continuously corrected. Outdoors, correlation is continuously performed using GPS reference and LiDAR sensors. The tracking of the respective position data via simple "state control" is interesting in comparison to model-based "predictive" forecasting – and finally, the AI-based synthesis of all incoming camera, sensor, and LiDAR data. It turns out that with only a single data source, deviations of up to several meters could indeed occur over longer distances.
Two 360° LiDAR systems for distance measurement, three image processing systems, and three safety LiDAR arrays for potential emergency braking are all working together. The edge server has its work cut out for it, processing this flood of data and using a program called YOLOv8, which Professor Konrad Doll aptly summarizes as "You only look once," as well as continuously retraining the AI software to produce precise readings. Fortunately, the developers also realized that it wasn't necessary to display every single detail, from screwdrivers to tiny mice, in their exact outlines. The primary focus, however, was on recognizing pedestrians (warehouse employees), other forklifts, gate outlines, ramps, and trucks.
Photos: Linde MH
Finally, there was even enough money for a sophisticated automatic battery charging system on the forklift, which uses a camera to position itself and a robotic arm to insert a plug into the charging socket. On the mast, the camera used to position the forks was even equipped with a lens cleaning system using spray nozzles.
In the outdoor area, a dummy demonstrates that the detection and numerous safety systems, interconnected via an edge server, can identify a (remotely controlled) wanderer in time, even in an awkward corner or behind a wooden wall. The forklift approaching the "unlucky" individual cannot see the employee from its position (because it is obscured). However, a second forklift has a clear view of the person, sends the data to the driver, and thus prompts him to brake. Even bystanders – although separated from the action by red barrier tape – appreciate this additional safety feature.
Author: K. Koch
www.linde-mh.de
