Toward a Clustering-Based Approach for Self-Adjusting Impact Factors in Robotic Control Model

In mobile robotic control models, control parameters are always generated by sensors’ information and a set of Impact Factors (IFs, such as the P-value in the PID model). The IFs take forms of fixed coefficients in control models and need to be pre-defined at design-time. However, when operating in an open environment, IFs of the control model are expected to be adjusted automatically at run-time in order to adapt to the environment changes and improve the operation of robotics. This paper presents a clustering-based approach to continuously updating the IFs in robot control model. The proposed approach utilizes the density-based clustering method to classify environmental changes based on the effects of these changes on robots. In each cluster, the regression method is designed to learn the relationship between IFs and environment changes, and therefore generate corresponding IF adjustment model. Such approach can decrease the mutual interference of environmental changes and enhance the rationality of robotic actions. The paper presents the self-adjusting framework and designs corresponding IFs update algorithms. This paper develops robotics path-following scenario and object-following scenario in open environment and conducts experiments to evaluate the effectiveness of the proposed approach. The results show that the proposed approach has faster response to environmental changes than DQN and MPC approaches, along with a lower deviation of robot’s actions.