Context: In the age of rapid e-commerce growth; the Robotic Mobile Fulfillment Systems (RMFS) have become the major trend in warehouse automation. These systems involve the use of self- governed mobile chares to collect shelves as well as orders for deliveries with regard to optimization of task allocation and with reduced expenses. However, in a manner to implement such systems, one needs to find enhanced algorithms pertaining to resource mapping and the planning of movement of robots in sensitive environments. Problem Statement: Despite RMFS have certain challenges especially when it comes to the distribution of tasks and the overall distances that employees have to cover. Objective: The main goal of this paper is to propose a new compound optimization model based on RL-ACO to optimize the RMFS’s task assignment and navigation. Also, the direction of the study is to investigate how such methods can be applied to real-life warehouse automation and how effective such methods can be on a large scale. Methodology: This research introduces a new optimization model for RMFS selection which integrates reinforcement learning with Ant Colony Optimization (ACO). Specifically, a real gym environment was created to perform the order assignment and training in the way of robotic movement. Reinforcement Learning (RL) models were trained with Proximal Policy Optimization (PPO) for improving the dynamic control of robots and ACO was used for computing optimal shelf trajectories. The performance was also measured by policy gradient loss, travelled distance and time taken to complete the tasks. Results: The proposed framework showed potential in enhancing the efficiency of tasks required and the travel distances involved. In each of the RL models used the shortest paths were identified and the best route was determined to have a total distance of 102.91 units. Also, other values such as, value function loss and policy gradient loss showed learning and convergence in iterations. To build a global solution, ACO integration went a step forward in enabling route optimization through effective combinatorial problems solving. Implications: This research offers a practical, generalizable and flexible approach for the improvement of the operations of RMFS and thinking for warehouse automation.
