Full State Estimation of Soft Robots From Tip Velocities: A Cosserat-Theoretic Boundary Observer

Soft robotics is a rapidly growing research area in robotics. Sensing of robotic systems is important for designing feedback controllers which usually provide robustness to modeling uncertainties. Sensing of soft robots, however, is considered a challenging task because theoretically, soft robots have infinite degrees of freedom while existing sensors only provide a limited number of measurements. One solution to this challenge is to design an observer/filter to estimate the unknown states from the sensor measurements. In this work, we design a boundary observer for soft robots based on the well-known Cosserat rod theory which models soft robots by nonlinear partial differential equations (PDEs). This boundary observer is able to estimate all the continuous robot states (poses, strains, and velocities) using the PDE model, inputs, and only tip velocity measurements (which explains the name "boundary" observer). The key idea is to inject sequential tip velocity measurements into the observer in a way that dissipates the energy of state estimation errors through the boundary. This boundary observer only requires sensing the tip velocity, can be implemented by simply changing a boundary condition in any numerical solvers of Cosserat rod models, and is proven to be locally input-to-state stable. Simulation studies are included and suggest that the domain of attraction is large and the observer is robust to measurement noise.