Control Loop

This is the perception phase of the loop. Raw data from sensors (e.g., encoders, IMUs, cameras) is read via interrupts or polling. This data is often noisy and incomplete, so it's fused using algorithms like a Kalman Filter to produce a coherent state estimate—a best guess of the system's current position, velocity, and orientation. The accuracy and latency of this estimate directly limit control performance.

The core algorithmic brain of the loop. It takes the desired setpoint (goal state) and the current estimated state to compute an error. A control law (e.g., PID, MPC) processes this error to determine the necessary corrective output.

• Proportional (P): Responds to current error.
• Integral (I): Accumulates past error to eliminate steady-state offset.
• Derivative (D): Predicts future error based on its rate of change.
• Model Predictive Control (MPC): Solves an optimization over a future time horizon for complex, constrained systems.

The physical actuation phase. The computed control signal (e.g., a voltage or torque command) is sent to the system's actuators—motors, servos, or hydraulic valves. This stage involves digital-to-analog conversion and often lower-level drivers like Pulse-Width Modulation (PWM) or Field-Oriented Control (FOC) for motors. The fidelity and bandwidth of the actuator determine how accurately the computed command is realized in the physical world.

The most critical real-time parameter. The loop period is the fixed, guaranteed time interval for one complete sensor-to-actuator cycle, often ranging from microseconds for motor control to milliseconds for higher-level planning. Deterministic execution means this period is jitter-free; missing a deadline (hard real-time constraint) can cause instability or failure. This is enforced by a Real-Time Operating System (RTOS) using Rate-Monotonic or Earliest-Deadline-First scheduling.

The physical system being controlled (e.g., a robotic arm, a drone) is called the plant. Its dynamics (mass, inertia, friction) are what the control law must counteract. Disturbances are unmodeled external forces acting on the plant, such as wind pushing a drone or a payload shifting in a gripper. A robust control loop must reject these disturbances to maintain performance, which is a key reason for the integral term in PID or the feedback nature of the loop itself.

The mechanism that closes the loop, creating a feedback control system. The output of the plant (its new state) is measured by the sensors, which feeds back into the controller as the new input. This continuous comparison between desired and actual state allows the system to correct for errors and disturbances in real-time. The alternative, open-loop control, sends pre-computed commands without feedback, which is ineffective for dynamic, uncertain environments.

A six-axis robotic arm performing a pick-and-place operation on an assembly line executes a high-speed control loop.

• Sensors: Encoders on each joint provide precise angular position feedback.
• Controller: A PID or model predictive control (MPC) algorithm calculates the torque required for each motor to follow a pre-planned trajectory.
• Actuators: Servo motors apply the computed torque.
• Loop Period: Typically 1-10 milliseconds. Missing a deadline can cause a collision or defective product.

An autonomous vehicle's lateral control (steering) is governed by a tightly coupled perception-planning-control loop.

• Perception: Cameras and LiDAR detect lane markings and the vehicle's offset from the lane center.
• Planning: A path planner generates a smooth, collision-free trajectory.
• Control: A steering controller (often MPC) calculates the exact steering angle needed to follow the path, accounting for vehicle dynamics.
• Actuation: A steer-by-wire system turns the wheels. This loop must run at 50-100 Hz to ensure safe handling at highway speeds.

A drone maintains stable hover through four independent motor speed control loops, managed by a central attitude controller.

• Sensors: An Inertial Measurement Unit (IMU) provides real-time angular rate and orientation data.
• State Estimation: A Kalman filter fuses IMU data to produce a clean estimate of roll, pitch, and yaw.
• Control: A PID controller compares the estimated attitude to the desired (often zero) attitude and computes corrective motor thrusts.
• Actuation: Electronic Speed Controllers (ESCs) adjust the RPM of each brushless motor. This loop runs at 400-1000 Hz to counteract instability.

A Computer Numerical Control (CNC) mill uses a multi-axis control loop to follow a toolpath with micron-level accuracy.

• Sensors: Linear scales and rotary encoders provide direct feedback on the position of the cutting tool relative to the workpiece.
• Control: The machine controller uses feedforward and feedback control to compensate for mechanical backlash and cutting forces in real-time.
• Actuators: High-torque servo motors drive ball screws to move the machine bed and spindle.
• Criticality: Loop jitter or latency directly translates to surface finish defects and dimensional inaccuracies in the machined part.

The hotend and heated bed of a 3D printer use dedicated thermal control loops to maintain precise temperatures.

• Sensor: A thermistor or thermocouple measures the current temperature.
• Controller: A PID controller computes the power (PWM duty cycle) needed for the heating element to reach and hold the setpoint.
• Actuator: A MOSFET switches current to the heater.
• Challenge: The system has significant thermal inertia and must reject disturbances from cooling fans and ambient temperature changes. Poor tuning leads to extrusion issues and print failures.

A humanoid robot maintains dynamic balance through a hierarchical control loop architecture.

• High-Level (Planning): Generates footstep locations and a center-of-mass trajectory at ~10-30 Hz.
• Mid-Level (Whole-Body Control): Solves for joint torques to achieve the planned motion while respecting balance constraints at ~100-500 Hz.
• Low-Level (Joint Control): Torque-controlled or position-controlled loops at each joint execute the commanded torques/positions at ~1-5 kHz.
• Feedback: Force-Torque sensors in the feet and IMU data are critical for detecting and correcting imbalance. This is a canonical example of nested control loops.

Proportional-Integral-Derivative (PID) control is a ubiquitous feedback control algorithm implemented within a control loop. It calculates an error value (e) as the difference between a desired setpoint and a measured process variable, then applies a correction using three terms:

• Proportional (P): Reacts to the current error.
• Integral (I): Accumulates past errors to eliminate steady-state offset.
• Derivative (D): Predicts future error based on its rate of change. The tuned gains (Kp, Ki, Kd) determine the system's response. It is the workhorse algorithm for stabilizing robotic joints, drone altitude, and thermal systems.

Model Predictive Control (MPC) is an advanced control method where the control loop uses an internal dynamic model of the system (e.g., robot dynamics) to predict future states over a finite time horizon. At each control cycle, it solves an online optimization problem to find a sequence of control inputs that minimizes a cost function (e.g., tracking error, energy use) while respecting constraints (e.g., actuator limits, obstacles). Only the first input is applied before re-solving at the next step. MPC is computationally intensive but provides superior performance for complex, constrained systems like autonomous vehicles and legged robots.

Jitter is the variation in latency or the deviation from the true periodicity of a periodic signal. In a control loop, it manifests as uneven time intervals between successive loop executions. Sources include:

• Non-deterministic OS scheduling.
• Variable interrupt service routine (ISR) latency.
• Garbage collection pauses in non-real-time languages. Excessive jitter degrades control performance, can destabilize feedback systems, and violates hard real-time guarantees. Mitigation involves using an RTOS, careful scheduling, and Direct Memory Access (DMA) for predictable I/O.

Worst-Case Execution Time (WCET) is the maximum possible time a specific task or block of code (e.g., one iteration of a control algorithm) can take to execute on given hardware under all possible inputs and states. It is a critical parameter for schedulability analysis in real-time systems. Determining an accurate, safe WCET is complex, involving:

• Static analysis of the code path.
• Hardware timing analysis (cache, pipeline effects).
• Measurement under extreme conditions. A control loop's period must be greater than the WCET of all tasks within it to guarantee deadlines are always met.