Reality arrives incomplete.
Sensors are noisy. Objects are occluded. Signals disagree. The same physical event can look different from another angle or under another condition. Perception is therefore not a perfect transcription of the world; it is an inference from partial evidence.
Fusion matters because context is distributed.
Depth can disambiguate an image. Motion can explain a sound. Location can change the meaning of a gesture. Time can distinguish a one-off anomaly from a pattern. Multimodal perception is less about collecting every signal than combining the signals that matter.
People are not objects with extra metadata.
Recognizing a person, pose or expression is different from understanding a human situation. Human meaning depends on relationships, history, norms and intentions that may not be directly observable.
Perception is the hinge between senses and sensitivity.
Below it are signals. Above it are judgments about meaning. That hinge is where technical sensing begins to meet the complexity of human context.