Because of the dynamic nature and lack of infrastructure of MANETs, routing in such networks is vulnerable to several attacks, and standard fixed policy routing algorithms are inefficient in tackling them. Reinforcement learning methods and appropriate models based on trust are assuring for dealing with the issues and variable behaviour of rogue network nodes. In this research, we present a cognition layer that interacts with the network layer in parallel and consists of two phases: pathfinding (routing) and trust assessment.The first phase uses ML techniques, while the second is focused on trust assessment. Regarding three performance indicators, CTR, our technique is evaluated against a well-known protocol, TQR. The simulation findings demonstrate improved end-to-end latency and communication.