A Study on Position Control of a Continuum Arm Using MAML (Model-Agnostic Meta-Learning) for Adapting Different Loading Conditions

Abstract

Predicting tip positions of a spring based continuum manipulator is highly challenging due to its nonlinear deformations. External loading on the tip further deteriorates the accuracy. Model-less control strategies have shown great success in the tip positioning. However, they require a large amount of data and time for the training. Performances of these controllers also deteriorate with external loads. To address these problems, this paper presents a MAML(Model-Agnostic Meta-Learning) based closed loop controller for the continuum manipulators. This controller requires a relatively small amount of data to achieve the state-of-art positioning accuracy. It can also adapt to changes due to the external loads with less than 2.5 percent of the original data. Two algorithms for the offline adaptation of the known and unknown external loading are proposed here. These techniques are also helpful for automatic stiffness tuning of the spring based continuum manipulators. The experimental validations have been done both in the simulation environment and on the real prototype. The continuum arm used for the experimentation is a tendon based non constant curvature spring-based manipulator. The average relative positioning error for the zero loading case was found to be 3.83% on the spring based prototype. The controller was successful in bringing down the relative tip positioning error of the manipulator from 5.42% to 2.7% in the simulation environment. It also showed success in bringing down the relative tip positioning error from 7.8% to 3.96% on the real prototype. Average relative tip positioning errors below 4.27% and 4.89% have been achieved in the trajectory following tasks for the known and unknown external loading cases respectively.