In the field of autonomous driving, automatic parking stands out for its ability to enhance convenience, alleviate traffic congestion, and improve parking efficiency. Traditional parking technologies, relying on complex sensor integration and precise path planning, often fall short in unpredictable and intricate situations. Deep Reinforcement Learning (DRL) has emerged as a promising approach to address these challenges, yet its application in parking still faces limitations, such as a narrow focus on conventional parking methods and a lack of capturing the full unpredictability of real-world parking conditions. This research seeks to advance the field by utilizing the Highway_env repository to create realistic parking scenarios that include reverse, parallel, and advanced diagonal parking scenarios. These scenarios vary from open to obstructed environments, increasing the realism and complexity of the parking task. Additionally, this study integrates Convolutional Neural Networks (CNNs) with DRL to enhance parking precision in diverse settings. Through detailed analysis, the study reveals that while both Soft Actor-Critic (SAC) and Twin Delayed Deep Deterministic Policy Gradient (TD3) algorithms demonstrate high success rate in simple scenarios (up to 0.99 success rate), TD3 displays superior performance in complex environments by exhibiting greater adaptability and successfully acquiring safer parking strategies, achieving success rate levels between 0.8 and 0.93. This highlights TD3’s exceptional adaptability and resilience in tackling a wide range of parking challenges, establishing it as a key solution capable of navigating the complexities of automatic parking.