A Context Aware Framework For Collision Avoidance With Deep Reinforcement Learning in VANET

Abstract

VANET (Vehicular Ad Hoc Network) is the most prominent technology that helps in the enhancement of performance and safety of the transportation system. The ad hoc network established between the moving vehicles on the road defines the VANET. VANET is distinguished by the mobility and self-organization of nodes, vehicles are considered as nodes here.V2V (Vehicle-to-Vehicle), V2I (Vehicle-to-Infrastructure) communications use VANET to warn vehicles nearby about emergencies (i.e, traffic jams, accidents), usually of the order of a few meters.Latency, secure transmission of messages, high-reliability criteria are the key challenges that one faces when implementing VANET. The primarygoal of this paper is to predict the collision from the sensor readings and prevent the happening of collision and V2V communication using ESP-NOW protocol .ESP-NOW helps in the transmission of wireless messages between the vehicle nodes which are controlled by ESP32 microcontroller. Reinforcement Learning (RL) with deep policy gradient networks is chosen here. Policy gradient (PG) methods are widely used algorithms for RL.Using Deep Reinforcement Learning (DRL) on the ESP32, the car node maintains a safe distance from the nearest obstacle. ESP-NOW protocol helps in exchanging the information between vehicle nodes, by finding all the MAC addresses of vehicle nodes that are present in its range and communicates the information. The focus of the proposed system is on taking the appropriate action using the DRL algorithm to avoid the collision risk and enables V2V communication which helps to broadcast the vehicle’s information to the other vehicles that are present in the V2V protocol range to maintain collision-free environment.