Safe Estimation of Vehicle Side-slip for an Autonomous Heavy Vehicle
Examensarbete för masterexamen
Systems, control and mechatronics (MPSYS), MSc
Stabilizing the lateral movement of the vehicle is crucial for functional safety. The variables required to ensure lateral stability cannot directly be measured from the vehicle. Hence, data from sensors is coupled with the estimates from the state estimator to obtain reasonably good estimates with a fair amount of accuracy. This technique called Sensor Fusion, helps to iteratively eliminate any outliers or anomalies and obtain the desired estimates. In this thesis, three different models of state estimators were developed in MATLAB/Simulink and their results compared. Their accuracy was compared to select the best model which would estimate the side-slip correctly. Each of the models has been given bounds (i.e., a minimum and a maximum bound) for values of side-slip. These bounds were based on tire wear-and-tear and other tire parameters which would simulate the real-world experiences. So the main goal was to design the estimator such that the bounds would encompass the actual measured value from the sensors (also called as 'ground truth'). The first model used was based on the linear bicycle model given in lateral dynamics. This was basically a mathematical model with a given set of equations along with vehicle parameters and variables. For vehicles with simple configurations (i.e., passenger cars and rigid-body trucks) whose lateral behaviour is relatively simpler compared to multi-axle or multi-trailer vehicles, the linear bicycle model gives reasonably accurate estimates as long as the sideslip angle doesn't exceed 0.5 degrees . The second method used was the kinematics model which integrated the lateral acceleration to give lateral velocity, which in turn was used to compute side-slip. However, as is present with most integrators, this methods suffers from what is called as integration drift. The third and final model was a washout filter model which was effectively a combination of a high-pass (HP) and a low-pass (LP) filter. Lateral velocities from both the bicycle model as well as the kinematic model act as inputs for this filter. This filter retains decent accuracy even for high values of side-slip and is not prone to the errors which the above-mentioned models suffer from.
side-slip angle , state estimator , tractor-semitrailer , lateral stability