Radio networks are getting too complex for hand-tuned rules. This handbook shows working wireless engineers how to design, train, and evaluate deep reinforcement learning controllers for the problems of 6G radio access networks: radio resource optimization, network slicing, and random-access congestion. Rather than starting from theory, each chapter connects a learning problem to the tools RAN engineers already use — link budgets, schedulers, and planning workflows — then builds the DRL solution step by step. You will learn how reinforcement learning maps onto RAN optimization, how native AI interfaces expose control knobs in modern architectures, and how O-RAN RIC and PPO control loops close the loop in practice. A full chapter compares OpenRAN Gym with robotics simulators, so you know exactly what transfers and what does not. The appendices provide complete, runnable Python environments — RANOpt and RANslice PPO examples plus PRACH storm scenarios in PPO and DQN — giving you starting code you can extend the same day. Moving an agent into a RIC requires compatible telemetry, an implemented actuator, and measured closed-loop validation; this book tells you precisely what each of those demands. For the engineer who needs working DRL in the RAN — not just the math.