Visual explainer

The Sense–Think–Act Loop

How robots observe changing environments, plan actions, and correct for delays and disturbances in closed loops.

1. Why Open-Loop Control Fails

Open-loop execution blind to changes causes collision.

An open-loop robot executes commands blindly without sampling the physical world during motion. If the terrain changes, wheel traction slips, or an obstacle sits in its path, the machine cannot adjust. Physical embodiment requires continuous observation: an autonomous agent must sense changes while moving.

2. The Feedback Triad

The Sense-Think-Act loop binds computation to physical reality.

The sense–think–act loop forms the core architecture of robotics. Sensors observe raw physical signals; computing hardware estimates state and plans actions; actuators exert physical forces back onto the environment. Every iteration revises the internal model before issuing the next torque command.

3. The Sense Beat: Acquiring Evidence

Sensors sample distance to physical obstacles at discrete intervals.

In this reference setup, a mobile robot advances at $v = 1.0\text{ m/s}$ toward a barrier. At timestamp $t_0 = 0.00\text{ s}$ , its forward-facing range sensor emits a pulse and samples a distance of $z = 1.20\text{ m}$ . Sensors operate at discrete sampling intervals rather than continuously.

4. The Think Beat: Planning Under Latency

Thinking consumes time while the robot continues moving forward.

Computation takes time. During a perception and planning delay of $\Delta t = 0.40\text{ s}$ , the vehicle continues forward by $d_{\text{travel}} = v \cdot \Delta t = 1.0 \times 0.40 = 0.40\text{ m}$ . Clearance to the obstacle shrinks from $1.20\text{ m}$ to $0.80\text{ m}$ before software issues a single braking command.

5. The Act Beat: Motor Torques in Action

Actuators convert commands into physical braking forces.

At $t = 0.40\text{ s}$ , motor drivers apply negative torque to the wheels, inducing a deceleration of $a = 2.0\text{ m/s}^2$ . Idealized stopping distance under constant deceleration is:

$d_{\text{brake}} = \frac{v^2}{2a} = \frac{1.0^2}{2 \times 2.0} = 0.25\text{ m}$

Braking requires $\Delta t_{\text{brake}} = v / a = 0.50\text{ s}$ of physical contact.

6. The Clearance Budget

Stopping distance divides into travel during latency and physical braking.

Total stopping distance combines the reaction buffer and the braking span: $d_{\text{stop}} = d_{\text{travel}} + d_{\text{brake}} = 0.40 + 0.25 = 0.65\text{ m}$ . Latency consumes more distance ( $0.40\text{ m}$ ) than mechanical braking ( $0.25\text{ m}$ ). The robot comes to a complete rest with $1.20 - 0.65 = 0.55\text{ m}$ of safety margin remaining.

7. Hierarchical Loop Rates

Robots run multiple nested control loops at different update rates.

A single loop rate cannot handle both complex reasoning and motor stability. Robots nest multiple control loops: low-level current and torque regulation runs at $500\text{--}1000\text{ Hz}$ , local trajectory tracking runs at $50\text{--}100\text{ Hz}$ , and high-level deliberative path planning runs at $1\text{--}5\text{ Hz}$ .

8. Rejecting Physical Disturbances

Feedback continuously corrects for unexpected physical disturbances.

When an inclined ramp or slippery patch introduces unmodelled drag, vehicle speed sags from $1.0\text{ m/s}$ to $0.6\text{ m/s}$ . The feedback loop measures the error $e_v = v_d - v = 0.4\text{ m/s}$ and ramps motor torque up by $+3.5\text{ N}\cdot\text{m}$ , restoring target velocity within $80\text{ ms}$ .

9. Closed-Loop Stability

Stable feedback drives error to zero while latency risks oscillation.

Fresh measurements allow error to decay smoothly to zero without overshoot. However, if sensing or actuation incurs excessive phase lag, the controller applies corrective forces long after the peak displacement has passed. This delay causes ringing, overshoot, and kinematic instability.

10. Where It Breaks

When latency exceeds distance to obstacle, collision occurs before actuation begins.

Feedback control collapses when obstacle distance is smaller than the latency displacement. If an unexpected obstacle appears at $d = 0.30\text{ m}$ , the robot requires $\Delta t = 0.40\text{ s}$ ( $0.40\text{ m}$ of travel) before computing a stop command. Impact occurs at full speed at $t = 0.30\text{ s}$ , before actuators can engage.

The Quick Version

Open-loop execution ignores environmental changes and risks physical collision.
The feedback loop links physical state, sensing, computation, and actuation.
Sensors sample physical variables at discrete intervals and timestamps.
Computation delay lets the robot advance forward before issuing commands.
Ideal braking distance follows $d = v^2 / (2a)$ under constant deceleration.
Total stopping distance equals reaction travel plus mechanical braking distance.
Hierarchical control nests fast motor loops inside slower planning loops.
Feedback detects unmodelled friction and slopes to reject physical disturbances.
Excessive phase lag and latency turn corrective feedback into unstable oscillations.
Obstacles appearing within the latency buffer cause unavoidable collisions.