Accurate  localization of  robots  in  a  specific  environment often requires the cooperation of multiple sensors,  and how to establish a more general data fusion model is always a difficult problem. For the localization of robot fish in an indoor  pool  environment,  this  paper  proposes  an  adaptive fusional algorithm based on fuzzy inference of dynamic weights. This paper firstly constructs a confidence probability table of sensors’ signals based on the calibration data sets of BLE and UWB nodes at different distances, as the basis for updating the weights of the BLE or UWB nodes; secondly, the data obtained by  each node  in  a  single  sampling  period in  the  robot  fish movement is vectorized to form a judgment matrix, and then the estimated distance is obtained by fuzzy inference together with the sensor weight; finally, the coordinates are calculated by the four-point positioning method. In this paper, more than 40 sets of  experiments  have  been  carried  out  with  a  simplified carangidae-like robot fish. The results show that the average positioning error is about 0.189m, which is 88.3% and 31.8% lower than that of using BLE only and using UWB only. In this paper,  the  fusional positioning  method  based  on  statistics combines the advantages of different sensors to reduce the data scale and achieve data denoising while fusing sensor data, which provides a reference for indoor positioning of robot fish and multi-sensor data fusion.