Purpose/Significance To propose a reinforcement learning-integrated optimization framework to address the adaptabilityproblems of algorithms in unstructured medical environments，and to enhance the reliability of dynamic navigation for inspection robots. Method/Process The proximal policy optimization algorithm is adopted to refine heuristic search strategies，aiming to enhance the effi.ciency and environmental adaptability of branch and bound search. By utilizing state awareness and reward feedback，dynamic parameter interaction is achieved，and branch generation logic is optimized through hierarchical search strategies. Multi-dimensional memory opti.mization is implemented based on experience compression and model quantization techniques. Result/Conclusion The optimized algo. rithm achieves millimeter-level localization accuracy required in medical scenarios，significantly reduces peak CPU utilization，shortens mapping time，and improves map completeness. It can operate stably on embedded platforms for medical robots，providing technical sup. port for clinical inspection tasks.