Task Learning And Planning Of Robotic Lamb By Reinforcement Learning From Human Feedback

A smart service robot has enormous potential to facilitate services for humans in family lives.However,there are still some challenges for the robot especially in performing daily life tasks because of the complicated environment and the various demands of humans during daily activities,which make it essential for a service robot to hace capability of task planning and task learning.When serving,the robots not only have to complete the mission,but also have to execute it in a right way matching the demands of humans.To overcome the problems,we proposed a Task Planning and Learning method of Robotic Lamb by Reinforcement Learning from Human feedback,which not only can learn the tasks but also can learn how to execute the tasks.The main contributions of my work is as follows:1)Proposed a feedback-based task planning and learning paradigm for perception-enhanced robotic limb 2)Proposed a deep-learning based task model to describe a task with structured environment information,which can execute a task in different environments;3)Proposed the human-guided based task learning method using reinforcement learning for robotic limb,which uses human feedback to train a deep strategy neural network and generate a task model by combining the deep strategy neural network and other task information;4)Developed the method in a real robotic limb,and evaluated the method to prove that our method can learn the common daily life tasks of a service robot by a 50-task experiment.