Incremental learning for robots: A survey

Nowadays, the application scenarios of robots are constantly updated, and the amount of data is also growing. Traditional machine learning methods are difficult to adapt to the dynamic environment. Incremental learning technology simulates the human learning process, enabling robots to use old knowledge to speed up the learning of new tasks and learn new skills without forgetting old skills. Currently, there is still relatively little research on robot incremental learning. This paper mainly introduces the research progress of robot incremental learning. Firstly, a brief introduction to incremental learning is given. Secondly, from the perspective of parameters and models, this paper classifies the current mainstream methods of robot incremental learning into three categories: variable parameter methods, variable model methods and hybrid methods, which are discussed in details, separately. Furthermore, the corresponding application examples of incremental learning technology in the field of robotics are provided. Thirdly, the data sets and evaluation metrics commonly used in incremental learning are introduced. Finally, the future development trends are prospected. © 2024 Northeast University. All rights reserved.