Meta Reinforcement Learning (Meta-RL) is an emerging research area that combines reinforcement learning (RL) with meta-learning principles to enable agents to rapidly adapt to new tasks with limited experience. Unlike traditional RL, which requires extensive interaction with the environment for each task, Meta-RL trains agents to learn a prior or initialization that can generalize across tasks, allowing quick policy adaptation. Early approaches focused on optimization-based methods such as Model-Agnostic Meta-Learning (MAML) applied to RL, while recent research explores recurrent policy gradients, contextual policies, and memory-augmented neural networks for few-shot adaptation. Applications span robotics, autonomous driving, game AI, healthcare, and adaptive control systems, where task variability and data scarcity are critical challenges. Recent advances also investigate hierarchical meta-RL, multi-task adaptation, exploration strategies, model-based Meta-RL, and integration with deep neural architectures to improve sample efficiency, generalization, and robustness. Current studies further explore transfer learning, safe adaptation, and sim-to-real transfer, establishing Meta-RL as a promising paradigm for building flexible and intelligent agents in dynamic environments.