As a higher level of human cognition, emotion plays an important role in environment learning and understanding. In this paper, emotion is introduced into the robot search task. By combined with memory mechanism, a cognitive model is proposed, which is composed of seven parts: internal state, receptor, environmental state system, emotion system, dynamic knowledge base, behavior decision system and actuator. In further, the emotion system consists of three modules: Emotion generation, emotion state and emotion memory, where the emotion memory is used to provide internal rewards. The memory function is implemented in the dynamic knowledge base. Based on the theoretical framework of reinforcement learning, the learning algorithm of the cognitive model is designed by integrating emotional internal reward and memory as a new reward mechanism. The maze robot search task requiring “energy replenishment” is tested, and results show that when faced with different situations, the robot can generate different emotions. Combined with the previous memory, the robot's decision-making seems more “anthropomorphic”, which firstly proves the effectiveness of the design of the emotion and memory mechanism. The cognitive model, the cognitive model without emotion and the Q-learning algorithm based on the $\varepsilon$-greedy strategy are compared, and results show that with the mechanism of emotion and memory, the robot can learn faster, and at the same time, its learning process is more stable.