CA-CR: Context-Aware Congestion Resolution Method for Automated Guided Vehicles

Modern logistics solutions use Automated Guided Vehicles (AGVs) in order to streamline logistics tasks in various work environments. While handling logistics tasks, AGVs can encounter congestions in the form of traffic or blockages which can cause slow-downs and deadlocks. Congestions can occur randomly and unexpectedly in work environments that are unknown to the AGVs. In unknown environments, access to environmental information is limited. Therefore, avoiding or resolving a congestion is a difficult task for AGVs. In order to compensate for the lack of environmental information, the contextual information of the congestion can be used. This paper presents the Context-Aware Congestion Resolution (CACR) method for AGVs in unknown work environments. The method uses contextual information in the form of number of agents, the vehicles’ current battery levels, task priorities and agent positioning to resolve congestions through a multi-agent negotiation. The negotiation uses the contextual information to find various solutions to random and unpredictable congestion scenarios. The negotiations are aided with reinforcement learning to train AGVs to move towards a collective utility maximizing behavior. To test the AGVs performance, the trained behavior is compared to a baseline, non-trained behavior for evaluation of reinforcement learning’s impact on optimizing the resolution of congestions. Findings indicate that the trained behavior shows better scaling with number of agents, better battery utilization and collective utility compared to baseline.