This paper introduces a novel obstacle avoidance strategy for an underwater robot, synergistically integrating Hidden Markov Chains (HMC) with a Mamdani Fuzzy inference controller. The HMC forecasts a vector of future states, and the maximum a posteriori state probabilities derived from this prediction directly inform the control actions of the fuzzy system. The proposed fuzzy logic controller employs five input variables: the HMC’s predicted robot state (represented as crisp sets), instantaneous linear velocity, underwater depth, and yaw and pitch angular velocities. The fuzzy controller generates three output variables, which directly command the robot’s actuators: the propulsion motor, the yawing motor, and the ballasting system, the latter operating via an integrated hydraulic piston mechanism. Furthermore, this work develops dynamics-based models for the robot’s propulsion, steering, and ballasting subsystems, complemented by sensor fusion models designed to provide real-time control feedback. The paper also details the robot’s platform and its underlying system architecture, engineered to support multi-threaded, real-time control operations. Experimental and simulation results are presented to validate the efficacy of the proposed obstacle avoidance strategy, demonstrating its robustness in challenging underwater locomotion scenarios.