Autonomous Underwater Vehicles (AUVs) are indispensable in marine ecosystem monitoring, deep-sea environment exploration and environmental conservation tasks. The traditional AUV control system suffers from issues such as high-latency decision-making, limited computational resources and the necessity of robust navigation in dynamic under-water environments. This work introduces Adaptive Curiosity-Driven Proximal Policy Optimization (AC-PPO) within a Hybrid Edge-Cloud Architecture tailored for AUV operations. The proposed system integrates advanced reinforcement learning with edge computing to enhance real-time decision-making, exploration, and navigation efficiency. AC-PPO combines intrinsic curiosity-driven rewards for improved exploration and adaptive clipping mechanisms for stable policy optimization. Hybrid edge-cloud architecture is used for computationally distributing edge computing workloads in low-latency critical decisions. In contrast, cloud-based processing is used for complex analytics and model updates. AUV uses lightweight versions of deep learning models Efficient Net-Lite. The cloud layer is used for high-complexity tasks, including large-scale species identification and global environment mapping. The system was tested on features extracted from the underwater dataset. The proposed work achieves 97.8% object detection accuracy, 95.2% navigation precision, a 44% reduction in latency, a 40% improvement in area coverage and an 18% increase in energy efficiency compared to conventional methods.