Abstract:Cognition is a fundamental and key issue of intelligent robots, which can provide robots with intelligence in terms of various human cognition behaviors. In order to achieve the human’s high-level cognitive ability to complex environment and tasks, much effort has been made to the researches on neuroscience, cognitive science and related areas. Recently, biological research with information technology integration is an important trend of research. A series of computational neural models accounting for visual cognition process have been proposed. In particular, one of the remarkable results is the famous the hierarchical Max-Pooling (HMAX) model, which is a hierarchical feed-forward model of the ventral stream of the primate visual cortex. Each level in the model is designed according to the results of biological experiments, which mimics the structure and function of the visual cortex and can enhance the analysis ability in multiple visual recognition tasks. However, the HMAX model could extract position and scale invariance features with a feed-forward hierarchical architecture, but the lack of feedback limits its performance in pure classification tasks. On the other hand, artificial neural networks have achieved the best performance in many practical tasks to discover multiple levels of representations from data. Especially, by utilizing the deep neural networks, robots are able to have the ability of self learning and decision making.

The Development and Challenges of Robot Cognition