Comparison of Two System Identification Approaches for a Four-Wheel Differential Robot Based on Velocity Command...

Precise modeling of differential drive robots is crucial for effective control and trajectory planning in autonomous systems. A comparative analysis of two modeling approaches for a four-wheel differential drive robot is presented in this paper. The first approach, named Motor-Based Model (MBM), identifies four transfer functions, one for each motor, while the second approach, named Simplified Model (SM), uses only two transfer functions, one for linear velocity and another for angular velocity. Both models were validated by comparing their predicted trajectories against real odometry data obtained from a SLAM system implemented on a differential-drive robot. This provided a practical assessment of each model's accuracy and underscored the importance of model selection in control design and navigation tasks. The results showed that the Motor-Based Model (MBM) consistently outperformed the Simplified Model (SM) in terms of odometry accuracy, both in position and orientation. Across all trajectories, the average RMSE for position using MBM was 0.309 m, while the SM recorded a higher average RMSE of 0.414 m. Similarly, the maximum position error averaged 0.522 m for MBM and 0.710 m for SM, confirming that MBM is more accurate and consistent in position tracking. Regarding the results of orientation estimation, when averaged across all experiments, the MBM maintained a lower angular RMSE of 0.170 rad in contrast to SM, which achieves an RMSE of 0.239 rad. The maximum angular error was also higher for the MBM at 0.316 rad, compared to 0.447 rad for the SM. Moreover, the computational performance evaluation indicated that the SM consistently outperformed MBM, achieving a 30% reduction in simulation time and substantially lower memory usage. These results demonstrate the relationship between model complexity and accuracy and suggest that the motor-specific model is more appropriate for applications requiring precise mapping or localization, such as SLAM, while the simplified model may be suitable for simpler use cases with lower computational requirements, such as embedded systems with limited resources. This paper provides a practical evaluation of the accuracy and computational performance of two modeling approaches, highlighting the implications of model selection for the design of navigation tasks.