A mobile robot receives measurements, uses a program to decide and drives mechanisms. It then measures again. Repeating this loop allows correction when the floor, battery or position changes. Motors plus software do not imply environmental sensing: driving for two seconds differs from continuing until a measurement indicates that a line is centred.
Consider following a dark strip on a light background. Downward-facing sensors measure reflected light; software turns their readings into an estimate of the strip's position. If it appears to the robot's right, the decision can be to steer right by increasing the left side's motion relative to the right. The next measurement indicates whether that adjustment moved the robot towards or away from the strip.
This interpretation depends on assumptions: sensors are ordered as expected, contrast distinguishes the strip, motor direction is correct and the surface permits motion. Reversing sensor order makes a mathematically correct algorithm steer the wrong way. Spinning tracks without traction also fail to produce the expected displacement.
Investigate failures through three questions: does the measured value change with the scene? Does the calculated decision match that value? Does the mechanism respond to that decision? First inspect readings with motors disabled, then each side's direction with stable support and a clear area. Only then assess the combined system on a floor-level track, away from edges and people.
Autonomy does not imply artificial intelligence. An explicit rule can solve a bounded problem very well. AI models are useful when they add a needed capability and their failures can be recognised. Initially, explaining the robot's decision is more valuable than adding a camera without knowing how its output changes motion.