A Differentiable Dynamic Model for Musculoskeletal Simulation and Exoskeleton Control

Chao Hung Kuo, Jia Wei Chen, Yi Yang, Yu Hao Lan, Shao Wei Lu, Ching-Fu Wang, Yu Chun Lo, Chien-Lin Lin, Sheng Huang Lin*, Po Chuan Chen, You-Yin Chen*

*Corresponding author for this work

Research output: Contribution to journalArticlepeer-review

6 Scopus citations


An exoskeleton, a wearable device, was designed based on the user’s physical and cognitive interactions. The control of the exoskeleton uses biomedical signals reflecting the user intention as input, and its algorithm is calculated as an output to make the movement smooth. However, the process of transforming the input of biomedical signals, such as electromyography (EMG), into the output of adjusting the torque and angle of the exoskeleton is limited by a finite time lag and precision of trajectory prediction, which result in a mismatch between the subject and exoskeleton. Here, we propose an EMG-based single-joint exoskeleton system by merging a differentiable continuous system with a dynamic musculoskeletal model. The parameters of each muscle contraction were calculated and applied to the rigid exoskeleton system to predict the precise trajectory. The results revealed accurate torque and angle prediction for the knee exoskeleton and good performance of assistance during movement. Our method outperformed other models regarding the rate of convergence and execution time. In conclusion, a differentiable continuous system merged with a dynamic musculoskeletal model supported the effective and accurate performance of an exoskeleton controlled by EMG signals.
Original languageEnglish
Number of pages18
Issue number5
StatePublished - 9 May 2022


  • differentiable physics
  • electromyography (EMG)
  • musculoskeletal model
  • Hill-type muscle
  • exoskeleton
  • motor control
  • adjoint method
  • gradient
  • differential equation


Dive into the research topics of 'A Differentiable Dynamic Model for Musculoskeletal Simulation and Exoskeleton Control'. Together they form a unique fingerprint.

Cite this