Sensor fault tolerant sliding mode control using information filters with application to a two-wheeled mobile robot

This paper considers the problem of fault-tolerance in two wheeled mobile robots (2WMR), using a robust sliding mode controller (SMC) coupled with a fault detection and exclusion (FDE) method. SMC can handle external matched disturbances, unmodeled dynamics and partial loss of effectiveness of actuators. However, it can not deal with erroneous sensor measurements and total actuator faults. For this purpose, a multiple model-based FDE method based on an informational framework is used, in order to detect and isolate faulty sensors and actuators. The information form of the Kalman filter (IF) is used for state estimation. The proposed IF uses two prediction models: a mathematical model with SM control signals as inputs, and an odometric model that takes encoders measurements as inputs. Residuals are generated using the Kullback-Leibler (KL) divergence, which computes the dissimilarity between the probability distributions given by the predicted estimations and sensors measurements. The proposed strategy excludes erroneous sensors measurements from the estimation process, which makes the proposed SMC algorithm law tolerant to these faulty sensors. Total Actuator faults, are also detected and isolated, but since the controllability of the system is not preserved, the 2WMR is stopped and an alarm is generated along with the position of the robot. A real-time experimentation on a real differential mobile robot is performed to validate the proposed strategy.