Looking over the linked paper, the abstract includes "Both experimental and simulation results confirm the effectiveness of the achieved control algorithms...", however, I don't actually see any experimental results in the paper, just simulation. So they don't ever estimate the accuracy of the technique in real-world conditions that I can see.
There are ways to both calibrate odometry parameters and measure accuracy over time. I would do a google search for 'UMBmark method'. Here is an article explaining the UMBMark method. Using that method, you can make your wheel-encoder based dead-reckoning as accurate as possible.
You can improve dead-reckoning accuracy by combining inertial navigation with wheel encoders; inertial navigation is a good complement to wheel encoders because it can detect wheel slippage. But inertial navigation also accumulates unbounded errors, so while adding it can be an improvement, dead reckoning will still accrue unbounded errors overtime. You need some other means of localization to fix these errors periodically.
This paper compares various methods for visual (camera based) odometry and localization. Typically wheel-encoder based odometry is used for high frequency state propagation and camera-based localization can be used for lower frequency updates.