As the robotics industry accelerates toward the widespread deployment of autonomous vehicles, humanoid robots, industrial manipulators, and delivery systems, the necessity for a standardized, technically rigorous taxonomy of robot autonomy becomes paramount. Existing frameworks, notably the Society of Automotive Engineers (SAE) levels adopted by the National Highway Traffic Safety Administration (NHTSA), classify autonomy primarily through the lens of human attentiveness, operational domains, and fallback responsibilities. However, these taxonomies do not adequately address the underlying algorithmic control mechanisms, the transition from static training data to real-time learning, and the compounding computational architectures that define modern robotic systems. This report proposes a novel, five-tier framework for classifying robot control levels, shifting the paradigm from human-centric monitoring to machine-centric cognitive and control logic. The proposed levels progress from Level 0 (Operator Controlled) to Level 4 (Reinforcement Learning), where the robot utilizes deep reinforcement learning policies to navigate and adapt to dynamic environments without external supervision. Furthermore, the analysis explores the exponential scaling of computational power required at each progressive level and outlines the future trajectory of the industry through decentralized and federated reinforcement learning (FRL). By enabling robot fleets to share policy data and aggregate learning experiences through peer-to-peer networks and edge-cloud infrastructures, FRL presents a viable, scalable pathway to mitigating the severe computational constraints inherent in Level 4 autonomy.