Small language model-assisted zone prioritisation for UAV-integrated robotic mobile fulfilment systems

Abstract

Robotic mobile fulfilment systems (RMFS) and unmanned aerial vehicles (UAVs) are among the new solutions in handling goods and order fulfilment in high-tech warehouse logistics, which are on the rise. A major issue in the warehouses is the dynamic situation that is usually present in warehouses, causing unreliable task priorities for the different parts of the warehouse, leading to low scalability and poor resource rates. This paper proposes an innovative framework that helps to function the algorithms with the help of small language models (SLMs), developed for the edge deployment in RMFS. Customer order information is the SLMs’ main input, including the processing of natural language specifications on the item's urgency and requirements, which can be used to produce prioritised zone and uncertainty quantification. The zones obtained are further optimised based on the improved parallel multi-ant colony optimisation (IP-MACO) algorithm, which is made fair and efficient by our modifications for the UAV navigation. In the virtual warehouse model, simulations prove the effectiveness of this approach in cutting down fulfilment time, energy savings and also in achieving better overall task fairness. This method works on the conversed drawbacks posed by traditional heuristics, facilitating proper operations in dynamic and high-speed RMFS.