t—Reducing unexpected urban traffic congestioncaused by en-route events (e.g., road closures, car crashes,etc.) often requires fast and accurate reactions to choose thebest-fit traffic signals. Traditional traffic light control systems,such as SCATS and SCOOT, are not efficient as their trafficdata provided by induction loops has a low update frequency(i.e., longer than 1 minute). Moreover, the traffic light signalplans used by these systems are selected from a limited set ofcandidate plans pre-programmed prior to unexpected events’occurrence. Recent research demonstrates that camera-basedtraffic light systems controlled by deep reinforcement learning (DRL) algorithms are more effective in reducing trafficcongestion, in which the cameras can provide high-frequencyhigh-resolution traffic data. However, these systems are costlyto deploy in big cities due to the excessive potential upgradesrequired to road infrastructure. In this paper, we argue thatUnmanned Aerial Vehicles (UAVs) can play a crucial role indealing with unexpected traffic congestion because UAVs withonboard cameras can be economically deployed when and whereunexpected congestion occurs. Then, we propose a system called“AVARS” that explores the potential of using UAVs to reduceunexpected urban traffic congestion using DRL-based traffic lightsignal control. This approach is validated on a widely used opensource traffic simulator with practical UAV settings, including itstraffic monitoring ranges and battery lifetime. Our simulationresults show that AVARS can effectively recover the unexpectedtraffic congestion in Dublin, Ireland, back to its original uncongested level within the typical battery life duration of a UAV.