随着机器人与人类工作和生活联系的日趋紧密，使机器人完成复杂任务的需求愈加迫切。以手动编程的方式增加机器人功能的做法不仅任务日益繁重且往往难以满足不断变化的新需求，迫切需要机器人像人类同时具备模仿和自主学习的能力。目前的技能学习技术往往是单一的方法，未能建立不同学习方法之间的联系，无法充分利用不同方法实现优势互补，因此面对不同场景复杂变化的任务常常难以为计。本课题将在前期研究积累基础上，深入分析人类技能学习机理，以示教学习和强化学习为研究对象，研究将二者有机的结合机制，有效利用不同学习方法实现优势互补。拟提出一种基于示教学习和强化学习的机器人类人技能学习框架，该框架不仅可以利用于仿真环境学习，还支持机器人通过与实际环境交互进行学习。该框架可以有效提高机器人自主学习复杂任务的能力，同时大大节约机器人技能更新的工作量。本课题研究是对现有研究难点和不足的一种创新性探索。

As the task for the robot become more complicated, manually robot programming is more time consuming and inconvenient. The robot should be capable of learning using supervised and reinforcement learning like human. Traditional methods only employ single learning method, have not achieved improvement through combination of different learning approaches. The topic is to analyze the way how human-being learn skills and use the idea on the robot skill learning tasks. This work intents to combine different skill learning methods to take full advantages of each one. The reinforcement learning algorithm is initialized using the policies learned from demonstrations, then deep learning method is used to enable the reinforcement learning algorithm with the capability of generalization. It also helps to solve the dimension curse problem. The proposed algorithm can save lots of time and effort during updating the robot with new skills.