Space manipulators could promote space exploration by assisting astronauts in space missions. Hence, controlling and modelling for space manipulators has been attracting research interest. However, the control of space manipulators presents challenges because of the coupling dynamics effects, the uncertain parameters and unknown disturbances. In this thesis, 2 types of adaptive sliding mode controller able to handle the uncertainties are designed to control the motions of both the manipulator’s joints and the spacecraft base.
A radius basic function neural network–based less chattering sliding mode controller (NNLSMC) is designed. The neural network is used to approximate the lumped effects of uncertainties. The adaptive law of switching gain is positively proportional to the absolute value of sliding variables to achieve the convergence of sliding variables, while the adaptive law switches to the designed tangent function to decrease the chattering effects once the sliding variables move into the vicinity. The controller demonstrates the ability to mitigate chattering effects without the loss of robustness.
However, the NNLSMC indicates the possibility of decreasing accuracy. Although the loss of accuracy can be mitigated by appropriately setting the parameters of adaptive law, it is laborious to manually tune those parameters. Thus, a reinforcement learning–based adaptive sliding mode controller (RLFSMC) is proposed, which has a similar problem to NNLSMC (mitigation of chattering effects, yet loss of accuracy). However, the exact parameters of the adaptive law are determined by the fuzzy logic inference, in which the fuzzy rules are automatically determined by modified reinforcement Q learning. Therefore, RLFSMC is less labour intensive and has greater potential to find better parameters than does the NNLSMC.