摘要
提出一种基于栈式自动编码器(Stacked Auto Encoder,SAE)和长短时记忆(Long Short-Term Memory,LSTM)神经网络相结合的步态预测方法解决下肢外骨骼机器人跟随控制问题。人体在行走过程中下肢步态具有一定的周期性,通过将下肢运动信息作为输入,步态作为输出,构建SAE-LSTM神经网络模型,并利用Keras对SAELSTM神经网络进行搭建和验证。实验结果表明,SAE-LSTM神经网络根据之前时间段的步态序列有效地预测出下一时刻的步态信息,平均准确率能够达到92.9%以上。
A gait prediction method based on SAE and LSTM neural network is proposed to solve the problem of followup control of lower extremity exoskeleton robot. Since the human body has a certain periodicity of the lower limbs posture during walking, the SAE-LSTM neural network model is constructed, by using the lower extremity motion information as inputs, the gait as an output. And using the Keras to build and validate the SAE-LSTM neural network. The experimental results show that the SAE-LSTM neural network can effectively predict the gait information at the next moment according to the previous gait sequence, and the average accuracy can reach more than 92.9%.
引文
[1]Yan T,Cempini M,Oddo C M,et al.Review of assistive strategies in powered lower-limb orthoses and exoskeletons[J].Robotics&Autonomous Systems,2015,64:120-136.
[2]Wilcox M,Rathore A,Ramirez D Z M,et al.Muscular activity and physical interaction forces during lower limb exoskeleton use[J].Healthcare Technology Letters,2016,3(4):273-279.
[3]Huo W,Mohammed S,Amirat Y,et al.Active impedance control of a lower limb exoskeleton to assist sit-tostand movement[C]//IEEE International Conference on Robotics and Automation,2016:3530-3536.
[4]龙亿,杜志江,王伟东.基于人体运动意图卡尔曼预测的外骨骼机器人控制及实验[J].机器人,2015,37(3):304-309.
[5]丁峰,韩云鹏,顾承超,等.基于灰色理论的人体步态预测[J].计算机应用与软件,2017,34(10):223-226.
[6]芮挺,费建超,周遊,等.基于深度卷积神经网络的行人检测[J].计算机工程与应用,2016,52(13):162-166.
[7]Islam B,Baharudin Z,Nallagownden P.Development of chaotically improved meta-heuristics and modified BPneural network-based model for electrical energy demand prediction in smart grid[J].Neural Computing and Applications,2017,28(1):877-891.
[8]Yamada T,Murata S,Arie H,et al.Representation learning of logic words by an RNN:from word sequences to robot actions[J].Frontiers in Neurorobotics,2017,11:70.
[9]Lecun Y,Bengio Y,Hinton G.Deep learning[J].Nature,2015,521(7553):436.
[10]Rose J,Gamble J G.Human walking[M].3rd ed.Philadelphia,PA,USA:Lippincott Williams&Wilkins,2006.
[11]Li J,Shen B,Chew C M,et al.Novel functional taskbased gait assistance control of lower extremity assistive device for level walking[J].IEEE Transactions on Industrial Electronics,2016,63(2):1096-1106.
[12]Gehring J,Miao Y,Metze F,et al.Extracting deep bottleneck features using stacked auto-encoders[C]//IEEEInternational Conference on Acoustics,Speech and Signal Processing,2013:3377-3381.
[13]Pascanu R,Mikolov T,Bengio Y.On the difficulty of training recurrent neural networks[C]//30th International Conference on International Conference on Machine Learning.Atlanta,USA:IMLS,2013:1310-1318.
[14]Graves A.Long short-term memory[M]//Supervised sequence labelling with recurrent neural networks.Berlin Heidelberg:Springer,2012:37-45.
[15]Greff K,Srivastava R K,Koutník J,et al.LSTM:a search space odyssey[J].IEEE Transactions on Neural Networks&Learning Systems,2015,28(10):2222-2232.
[16]Srivastava N,Hinton G,Krizhevsky A,et al.Dropout:a simple way to prevent neural networks from overfitting[J].The Journal of Machine Learning Research,2014,15(1):1929-1958.
[17]Kinga D,Ba J.Adam:a method for stochastic optimization[C]//International Conference on Learning Representations,2015.