Sensor Fusion and Adaptive Predictive Control for Autonomous Tracked Robotic Systems in Industrial Cleaning Processes

This study proposes an advanced control and monitoring framework for autonomous tracked robotic systems operating in highly unpredictable industrial cleaning environments, such as large-scale photovoltaic plants and rough factory floors. Tracked locomotion offers superior traction on debris and inclined surfaces; however, it suffers from severe non-linearities caused by track slippage, varying ground friction coefficients, and non-holonomic constraints. To address these challenges within the scope of advanced process control, a dual-layer architecture combining an Extended Kalman Filter (EKF) for sensor fusion and an Adaptive Model Predictive Control (AMPC) loop was developed. The sensor fusion layer integrates high-frequency data from a 6-DOF Inertial Measurement Unit (IMU), high-resolution optical track encoders, and time-of-flight distance sensors to compute real-time kinematic states and dynamically estimate slip ratios. These estimated states feed into the AMPC layer, which continuously updates the prediction model matrices based on localized track-soil interactions. The optimization problem is solved onboard at a control frequency of 50 Hz using a quadratic programming solver. Experimental validation on a 5 kg industrial tracked cleaning robot prototype demonstrated that the proposed EKF-AMPC framework reduced the maximum cross-track error to 1.1 cm, even when transitioning between high-friction dry surfaces and low-friction wet surfaces. Compared to conventional non-adaptive controllers, the proposed system yields a 64% improvement in trajectory tracking precision and completely eliminates oscillatory behaviors during sharp turns. This integration establishes a robust and highly autonomous solution for industrial process automation, effectively minimizing mechanical wear and optimizing energy utilization.