Reinforcement Learning & Multimodal AI enabled AGVs in logistics

Warehouses are dynamic environments where different assets are continuously changing. Autonomous navigation nowadays often relies on floor markings and (static) prior maps which are created by driving around and scanning the surrounding environment. When augmented with vision technologies and AI algorithms, AGVs (Automated Guided Vehicles) can gain a spectacular amount of autonomy even in unmarked and unknown environments. In collaboration with researchers from University of Antwerp and KU Leuven, a proof of concept was made illustrating how Reinforcement Learning is used for automated inventory counting. Reinforcement learning methods learn through trial and error to interact with an unknown environment and hence, for safety reasons, this is often done in simulation. The step to the real environment is made safely by making use of the sensors on the AGVs.