Learning-Based Real-Time Shape and Force Estimation for Continuum Robots Using FBG-Embedded Actuation Cables

Continuum robots are widely used in endoluminal intervention and treatment due to their dexterity and compliance. However, it is challenging to achieve simultaneous shape and force estimation of continuum robots under the requirement of robot miniaturization. To address it, this article proposes a learning-based real-time tip force and shape estimation for continuum robots using distributed tensions and proximal displacements of actuation fibers. It only needs the tension-sensing fibers for the actuation and sensing without using additional sensors. First, each utilized actuation cable is an optical fiber with inscribed distributed fiber Bragg grating sensors on it to provide distributed tension sensing. Finally, a neural network-based method is proposed to estimate the external force at the tip using the wavelength shifts of the fiber Bragg grating sensors on the actuating fibers and cable length changes as inputs, also the kinetostatic model is established to reconstruct the robot's shape in real time. Then, experimental results show that the average static force estimation error is 6.2 mN and the average tip position error from shape estimation is 3 mm. Furthermore, an augmented reality platform was developed to overlay the real-time estimated external forces and robot shape onto two-dimensional video images from an external monocular camera. This work would provide a potential to promote the embodied intelligence of continuum robots with miniaturization and integration.