Interception of adversarial drones in urban spaces using a modified deep reinforcement learning approach

Abstract

The proliferation of drone technologies in urban spaces has increased the threat space to potential attacks to key assets and people. The early disruption of adversarial drones is crucial to ensure safety and smooth operation of transportation services. Previous approaches use target interception algorithms for well-known trajectory profiles such that the interception problem is reduced in solving a geometric task. However, the unpredictable nature of the decision-making of drones poses significant challenges to traditional interception algorithms. To this end, this paper proposes a tactical and dynamic path planning approach that maximises the interception probability of adversarial drones in presence of obstacles, whilst minimising the interception time. This is achieved by modifying a twin-delayed deep deterministic policy gradient (TD3) algorithm with an artificial potential field (APF) that increases the likelihood of success for the interception in a reduced time, whilst avoiding local minima issues. Extensive experiments are conducted to demonstrate the feasibility of the approach and future research directions.