The Implementation Of Nonlinear UKF Algorithm For The Motion Trajectory Decoding

Brain Machine Interfaces (BMIs) aim to build a connection between brain and external devices, so as to help patients with physical disabilities to complete many tasks, and improving the quality of their life.Suppose there is a research result comes from a weak or unreasonable system, it obviously cannot be a believable or scientific achievement. So, in other words, a good and stable experiment platform can be quite important to any of BMI research field. And also, the encoding and decoding algorithm is one of key elements in the whole BMI system. There are many research achievements are published in these two decades and they also proposed their own decoding model or algorithms. Those articles can serve like important references to this thesis. More and more researches show that the tuning properties of neurons are kind of nonlinearity, that is, the linear model cannot give enough description to the BMIs neuron characters. So, as a starting point, we believe that those algorithms which based on nonlinear model can give better BMI performance. In this thesis, we mainly try to find an algorithm which suit for the online BMI experiment and can also give well enough decoding result at the same time. And then, transplant the selected algorithm into our BMI platform to test it online performance. Therefore, the main work of this thesis can be concluded in to two parts, algorithm and platform system.In the algorithm field, this thesis reviewed many classic decoding algorithms, compared their performance and time complexity from the theory side. And through research and many tests in offline data, we compared 5 kinds of decoding algorithms and give the result that the Unscented Kalman Filter (UKF), which base on nonlinear model, can perform quite well both in the decoding results and running time. Through implement the UKF algorithm in our BMI system, we extended our system’s available algorithms choices and prove that the UKF nonlinear decoding algorithm can meet the validity and velocity demand of online experiment. And our BMI system can turn out better research result under the application of UKF algorithm.In the platform field, the research thesis base on the an original version of BMI system, which programs’data structures are quite mess and functions are not good enough to be qualified as an online system. We teased its structures and made it more modularization, and developed lots of new functions to let it can meet the demand of online obstacle avoidance experiment. Moreover, we connected the program system with the JACO robot arm and make it can real time cooperate subject’s movement, realized the function demand of monkey online control robot arm to do obstacle avoidance task.