Graduate School of Biomedical Engineering Graduate School of Engineering, Tohoku University, Sendai, Japan.
Artif Organs. 2011 Mar;35(3):270-4. doi: 10.1111/j.1525-1594.2011.01223.x.
The Feedback Error Learning controller was found to be applicable to functional electrical stimulation control of wrist joint movements in control with subjects and computer simulation tests in our previous studies. However, sinusoidal trajectories were only used for the target joint angles and the artificial neural network (ANN) was trained for each trajectory. In this study, focusing on two-point reaching movement, target trajectories were generated by the minimum jerk model. In computer simulation tests, ANNs trained with different number of target trajectories under the same total number of control iterations (50 control trials) were compared. The inverse dynamics model (IDM) of the controlled limb realized by the trained ANN decreased the output power of the feedback controller and improved tracking performance to unlearned target trajectories. The IDM performed most effectively when target trajectory was changed every one control trial during ANN training.
在我们之前的研究中，发现反馈误差学习控制器可适用于腕关节运动的功能性电刺激控制，且经过了受试者控制和计算机模拟测试。然而，在之前的研究中，仅使用正弦轨迹作为目标关节角度，并且针对每个轨迹对人工神经网络 (ANN) 进行了训练。在本研究中，重点关注两点到达运动，通过最小冲量模型生成目标轨迹。在计算机模拟测试中，比较了在相同的总控制迭代次数（50 个控制试验）下，使用不同数量的目标轨迹训练的 ANN。通过训练后的 ANN 实现的受控肢体的逆动力学模型 (IDM) 降低了反馈控制器的输出功率，并提高了对未学习目标轨迹的跟踪性能。当在 ANN 训练过程中每进行一次控制试验就改变一次目标轨迹时，IDM 表现最为有效。
