Short-range Robotic Navigation And Exploration Tasks Via Deep Q-networks For Biomedical Applications

The performance of different deep reinforcement learning algorithms in the medical application-virtual operating room is our study in this paper.Through observation and analysis,we established a virtual environment system innovatively,which aims to explore how agents can take a series of actions to achieve their goals in this environment.The virtual system designed a punishment-reward mechanism to guide the agent to take the behavior of getting the maximum reward.Existing methods such as Simultaneous Localization and Mapping(SLAM)require input information to navigate to its goal target and may not complete its task when the agent collides with an obstacle.This method may not be useful for efficient workflow in an operating room setting.This paper reviews the influence of deep reinforcement learning algorithms on an agent in our innovative virtual environment for biomedical applications.In this paper,we compare the performance of different model-free algorithms in the simulated environment with other traditional methods.We performed experiments and analysis based on the learning rate,cumulative positive reward,and the average time taken to attains its goal position.We first train our agent in four different environments and validate our training in our innovative operating room.Our research aims to train an agent to make intelligent decisions in achieving its goal destination without prior experience and input data.Reinforcement learning provides a structure for robotics to function effectively;utilizing and engaging a robot to navigate and explore in any given environment.