Reacting vs. Deliberating

If you touch a hot stove, you pull your hand away before you even consciously think about it. That reflex is fast because it bypasses slow deliberate thought. Your brain uses two systems: one fast and automatic, one slow and careful. Robots face the exact same design tension. Should a robot react instantly to what its sensors detect, or should it think carefully before acting? The answer, as in biology, is usually both — but knowing when to use each is the real engineering challenge.

Reactive Architectures: Speed First

A reactive architecture is one where the robot's actions are driven directly by sensor inputs, with little or no internal planning. The famous subsumption architecture, designed by roboticist Rodney Brooks in the 1980s, organized behavior as a stack of simple rules: avoid obstacles, then wander, then follow a wall. Higher-level behaviors could override lower ones, but none of them involved constructing a world model or planning ahead. Reactive systems are fast because they skip the perceive-model-plan loop. A robot running a reactive obstacle-avoidance rule can respond in milliseconds because it only needs to read a sensor and apply a rule — it does not need to update a map, run a search algorithm, or calculate expected values. In fast-changing environments, this speed can be the difference between avoiding a collision and crashing.

Deliberative Architectures: Intelligence First

A deliberative architecture builds a detailed world model, plans multiple steps ahead, and only then acts. This is the classical AI approach — reason through the problem completely before committing to an action. Deliberative systems can produce much smarter, more globally optimal behavior than reactive ones. A deliberative robot navigating a building can plan the most efficient route through every room rather than just bumbling around avoiding walls. The cost is time. Building and searching a world model takes computation. In the seconds it takes to plan, the world may have changed. A deliberative robot that is too slow to respond to a fast-moving obstacle is dangerous.

Hybrid Architectures: The Best of Both

Most modern robots use hybrid architectures that layer reactive and deliberative components. A typical design has three tiers. The reactive tier handles immediate safety: if a sensor detects an imminent collision, the robot brakes or swerves within milliseconds, no plan required. The deliberative tier runs in the background, continuously planning routes and updating goals. A middle tier (sometimes called the sequencing tier) coordinates between the two: it sends high-level goals to the deliberative planner and monitors the reactive tier to make sure short-term safety behaviors do not derail the long-term mission. Self-driving cars exemplify this structure: reflexive braking when a child steps off a curb (reactive), route calculation to the destination (deliberative), and a coordinating layer that manages the transitions between highway cruise control, urban intersection behavior, and parking maneuvers.