ENVIRONMENT FORCE ESTIMATION WITH STOCHASTIC NOISE AND CONTROL FOR ROBOTICS

Abstract

The proposed work entitled “Environment Force Estimation with Stochastic Noise and Control for Robotics” focuses on estimation of state vector and controller’s parameters and control of a robot in the presence of noise. Also, environment force is estimated when robot has interaction with randomly structured environment. In the present work, two typical issues are investigated for the robotic system. The first one deals with the noise present in the system. Noise is wiped out from the output measurement by utilizing state-observer which in turn, feeds the estimated state to a controller in spite of noisy state. Thus, any further randomness is restricted and focus of the work is to design, implement and test the optimal state-observer-based controller. It aims for trajectory control of the state of an n-link robot to track the desired trajectory in the presence of stochastic noise. The novel feature of the present control algorithm is based on Itˆ o0s stochastic calculus. It is used for the minimization of the conditional expectation of the instantaneous tracking error energy differential with respect to the feedback matrix. This instantaneous minimization results in an adaptive action of the controller. At the same time, it also maintains the energy constraints in the feedback coefficients to avoid actuator saturation. This enables to validate the real-time implementation of state-observer-based controller by an experimental set up of “Phantom Omni Bundle” robot manipulator. Furthermore, the sensitivity analysis of controller gain variation and parameter variation on the error process is performed, and the robustness of the system is assured by the bounded errors. Another important and relevant issue is that when noise appears in the form of environmental torque and randomness becomes a part of the closed-loop system and makes the dynamics of the nonlinear system ‘stochastic’. Utilizing the sample data of noisy measurement of joint position, the ‘likelihood function’ of tracking error is computed. Minimization of this function estimates the unknown controller gain parameters. The trajectory of stochastic environment force is also computed iii from the estimates of controller gain parameters. Subsequently, the performance of estimation is determined using convergence analysis and imposing lower bound on the variance of estimation errors. In the present study, both issues are addressed for a nonlinear dynamical system in a stochastic scenario. Controllers are designed based on estimation without using any velocity sensor or force sensor.