Solving Real-Time Distribution of Pickup and Delivery Tasks with Multiple Robotic Agents

The advancement of automation technology has driven productivity and quality in industrial and logistics processes, reducing operational costs. In automated environments, such as warehouses and ports, fleets of robots continuously move loads without human intervention. The Multi-agent Pickup and Delivery (MAPD) problem involves multiple agents that attend to a continuous flow of pickup and delivery tasks in a known environment. Tasks arrive and are assigned to idle agents, which must move --free from collisions-- from their current positions to the pickup location and then to the delivery location. Considering the combinatorial nature of the problem and the need for real-time responses, we propose three scalable heuristics: Nearest Pickup (NP), Threshold Task Path (TTP), and Split Delivery Task (SDT). The proposed heuristics were tested on instances from the literature with up to 500 agents and 1000 tasks, and compared to state-of-the-art approaches. NP performed better in terms of computational time, while TTP and SDT concerning the path quality. The use of path quality goals in TTP and SDT resulted in improvements of 3\% to 10\% compared to NP, with an increase in computational cost of 10\% to 500\%, respectively.