In this paper, we propose a new variant of fish school search algorithm, called rate-learning based fish school search algorithm (RL-FSSA), that uses a new refinement strategy to guide the population of fishes towards the best solutions. The fish motion is based on a collective sensorimotor behavior. Their learning ability determines the fish with the best position (i.e., leader). In each cycle, the best fish ever found is refined by a rate learning based process. This refinement allows a local search and decreases the effect of the hazard through time, and the whole algorithm goes from diversification towards intensification. In order to evaluate its performance, the proposed algorithm is tested under a set of benchmark functions. The numerical tests include unimodal and multimodal functions. The results show the high performance of RL-FSSA compared to the standard version FSSA. Furthermore, the computational time of the fish school seach algorithm is reduced. For the validation, the algorithm is used for the trajectory planning and control of the mobile robot in an environment containing obstacles.