Reinforcement Learning (RL) is more than just an algorithm; it is a computational approach to learning from interaction. Unlike supervised learning, where a teacher provides the "correct" answer, RL is centered on a goal-directed agent that must discover which actions yield the most reward through trial and error.

The Sensorimotor Connection

Learning is grounded in a direct sensorimotor connection to the environment. Imagine a robotic arm learning to pick up a fragile object:

• Sensation: The robot perceives the object's position through cameras (state).
• Action: The robot moves its joints (influence).
• Goal: To lift the object without breaking it (the success metric).

Every choice changes the state of the world the agent is sensing. The robot strengthens the connections that lead to a successful grasp without needing a human to program every millimeter of movement.