Learning nonlinear dynamic models of soft robots for model predictive control with neural networks

Soft robots have the potential to significantly change the way that robots interact with the environment and with humans. However, accurately modeling soft robot dynamics in order to do model-based control is extremely time consuming and difficult. neural networks are a powerful tool for modeling systems with complex dynamics such as an inflatable robot link with antagonistic pneumatic actuation. Unfortunately it is also difficult to apply standard model-based control techniques using a neural net. In this work, we show that the gradients used within a neural net to relate system states and inputs to outputs can be used to formulate a linearized discrete state space representation of the system. Using the state space representation, model predictive control can be developed with a one degree of freedom soft robot to achieve position control within 2° of the commanded joint angle. Additionally, control using the model derived from the neural net has similar performance to control using a model derived from first principles that took significantly longer to develop. This shows the potential of combining empirical modeling approaches with model-based control for soft robots.