Development of An Inverse Dynamic Model

摘要: In a controller design process, it is interesting to find the inverse model, in which desired input signal of system determined by using output system. many cases, model problem very difficult, and sometimes impossible determine implement. There are some methods make an dynamic such as computed torque control, had been studied previous Chapters. The capability NNs learn plant has investigated for years; NN can be used approximate this approach, should known, trained obtain model. early studies adaptive learning control Barto et al. [1], Jordan [2], Miller [3] Psaltis [4] addressed how error training controller. Generally, cost function, consisting squared norm reference errors, not correctly train Therefore, [2] proposed forward-inverse-modeling, Albus [5], Atkeson Reinkensmeyer [6], [4], Kuperstein Rubinstein [7] direct-inverse-modeling command-error forming feedforward Moreover, Watanabe [8] linear with two layers unit function output-layer. Kawato [9] method controller, uses feedback conventional configured parallel same manner reported Gomi [10], Miyamoto [11], but different SF algorithm. another work, Wada [12]