Sensors and Actuators Guide: LiDAR, Cameras, Motors, Servos
Every robot interacts with the physical world through sensors and actuators. Sensors transduce physical phenomena — distance, light, acceleration, position — into electrical signals. Actuators convert electrical signals into physical motion. The choice and integration of these components determines what a robot can perceive and what it can accomplish. This guide provides a comprehensive reference for the sensors and actuators used in robotics, covering operating principles, interface protocols, performance characteristics, selection criteria, and sensor fusion techniques.
Sensor Fundamentals
Robot sensors are categorized by what they measure. Proprioceptive sensors measure the robot’s internal state — joint angles, wheel rotation, motor current. Exteroceptive sensors measure the external environment — distance to obstacles, visual appearance, temperature. The combination of both types enables the robot to understand its own state within its environment.
Key sensor specifications include resolution (the smallest detectable change), accuracy (closeness to true value), precision (repeatability of measurements), bandwidth (maximum update rate), and range (minimum and maximum measurable values). Understanding these specifications and their tradeoffs is essential for selecting the right sensor for a given application. Aliasing is a critical consideration when sampling sensor data. The Nyquist-Shannon sampling theorem requires a sampling rate at least twice the highest frequency component in the measured signal.
Everett’s comprehensive reference on sensor fundamentals for mobile robots provides detailed treatment of sensor physics, interface design, and trade-off analysis that remains relevant to modern robotics (Everett, Sensors for Mobile Robots: Theory and Application, A K Peters, 1995).
LiDAR Sensors
Light Detection and Ranging (LiDAR) sensors measure distance by emitting laser pulses and measuring the time of flight. LiDAR is the primary sensor for autonomous navigation due to its direct distance measurement, wide field of view, and robustness to lighting conditions. 2D LiDAR sensors scan a single plane. The RPLidar A1, widely used in educational and research robotics, scans 360 degrees at 5-10 Hz with a 12-meter range and 1-centimeter resolution. 2D LiDAR is sufficient for flat-floor navigation and is commonly used with the ROS navigation stack.
3D LiDAR sensors scan multiple planes to produce dense 3D point clouds. Solid-state LiDAR sensors — the Ouster OS series, Livox Mid-40 — use MEMS mirror scanning or flash illumination, reducing moving parts and cost compared to earlier spinning LiDAR designs. LiDAR selection considers range, field of view, point rate, accuracy, and cost. Automotive-grade 3D LiDAR ranges 100-300 meters at centimeter accuracy with prices from $1,000 to $10,000+ for high-end units.
Industrial safety-rated LiDAR — SICK microScan3 — includes integrated safety functions certified to SIL 3, enabling use as a primary safety sensor in autonomous vehicle applications. These sensors include configurable warning and protection zones with redundant output circuits.
Camera and Vision Sensors
Cameras provide dense visual information unmatched by any other sensor. A typical robot camera captures megapixel images at 30-90 frames per second. The global shutter vs rolling shutter distinction is critical for robotics. Global shutter sensors capture the entire frame simultaneously, essential for fast-moving robots where rolling shutter would distort moving objects. Rolling shutter sensors are cheaper and have better low-light performance but introduce artifacts during motion.
Stereo vision computes depth by triangulating features between two cameras with a known baseline. The ZED 2i and Intel RealSense D435 are popular stereo cameras for robotics, providing depth maps at up to 90 fps within a 0.2-20 meter range. Stereo depth accuracy degrades quadratically with distance — beyond 10 meters, the disparity is sub-pixel and depth estimates become unreliable.
Time-of-flight (ToF) cameras, such as the VL53L1X and the Microsoft Kinect, measure depth by illuminating the scene with modulated near-infrared light and measuring phase shift or pulse arrival time. ToF sensors provide dense depth maps at short range (up to 10 meters) with good accuracy regardless of texture, making them suitable for indoor manipulation and gesture recognition.
Event cameras (neuromorphic sensors) report per-pixel brightness changes asynchronously at microsecond resolution. They excel in high-speed and high-dynamic-range scenarios where conventional cameras fail — drones flying through trees, robots operating in rapidly changing lighting. The DVXplorer and Prophesee sensors are increasingly adopted in robotics research for high-speed control loops.
IMUs and Encoders
Inertial measurement units (IMUs) combine accelerometers and gyroscopes. A 6-axis IMU provides three-axis acceleration and three-axis angular velocity. A 9-axis IMU adds a magnetometer for heading reference. IMU selection considers noise density, bias stability, and bandwidth. Consumer IMUs (MPU6050, ICM-20948) cost under $10 but have bias stability of 50-100 degrees per hour. Industrial IMUs (ADIS16470, Epson M-G370) cost $200-500 with bias stability under 1 degree per hour. Navigation-grade IMUs cost thousands and provide sub-0.01 degree per hour stability.
Sensor fusion algorithms combine IMU data with other sensors. The Madgwick filter and Mahony filter use complementary filtering computationally efficient for embedded implementation. The extended Kalman filter provides optimal fusion at higher computational cost. Madgwick’s filter, published in 2010, became the de facto standard for drone attitude estimation due to its low computational cost and tunable gain parameter (Madgwick, Harrison, and Vaidyanathan, “Estimation of IMU and MARG Orientation Using a Gradient Descent Algorithm,” IEEE ICORR, 2011).
Quadrature encoders measure rotational position and direction. Incremental encoders produce A and B channel pulses with 90-degree phase offset. Absolute encoders output position directly over a digital interface (SSI, BiSS, or CANopen). Resolution ranges from 100 CPR (cheap hobby encoders) to 2^24 counts per revolution (high-end industrial absolute encoders with 16,777,216 counts per revolution). Magnetic encoders (AS5048, MA730) provide absolute position sensing without contacting the shaft, eliminating wear concerns in high-cycle applications.
Force and Tactile Sensing
Force-torque (F/T) sensors measure the six-axis forces and torques applied to a robot’s end-effector. These sensors are critical for assembly, polishing, and human-robot interaction. Six-axis F/T sensors use strain gauges arranged on a precision machined structure, with the signal processing computing forces and torques from the strain gauge bridge outputs. ATI Industrial Automation and Robotiq are leading F/T sensor manufacturers.
Tactile sensors provide contact location and pressure distribution across a surface. Resistive, capacitive, and piezoresistive tactile arrays with resolutions from 1 mm to 5 mm are available. The SynTouch BioTac provides human-like tactile sensing with a fluid-filled skin covering an array of impedance electrodes, plus a hydrophone for texture vibration sensing. Tactile sensing enables advanced manipulation capabilities — detecting slippage during grasping, identifying object texture, and performing precision assembly.
DC Motors and Servos
DC motors convert electrical energy into rotational mechanical energy. A permanent magnet DC motor’s speed is proportional to applied voltage, torque is proportional to current, and the torque-speed curve is approximately linear. Motor selection requires specification of nominal voltage, no-load speed, stall torque, and maximum permissible current. The motor’s mechanical time constant determines how quickly it responds to voltage changes.
Brushless DC (BLDC) motors replace the mechanical commutator with electronic commutation. BLDC motors offer higher efficiency, longer life, and better thermal performance than brushed motors. The tradeoff is more complex drive electronics requiring six MOSFETs with gate drivers and rotor position sensing. BLDC motor selection considers the KV rating (RPM per volt), stator resistance, and torque constant. High-KV motors are suited for high-speed, low-torque applications; low-KV motors provide high torque at lower speeds.
Servo motors integrate a DC motor, gearhead, position sensor, and control electronics into a single package. Hobby servos accept PWM position commands at 50 Hz with pulse widths mapping to 0-180 degree position. Smart servos (Dynamixel, Herkulex) accept digital commands over half-duplex UART or TTL serial, reporting position, velocity, temperature, and load. Dynamixel servos, widely used in research robotics, provide daisy-chain communication over a single bus, enabling simplified wiring for multi-jointed robots.
Soft and Unconventional Actuators
Soft robotics uses compliant actuators made from elastomers, textiles, or shape-memory alloys. Pneumatic artificial muscles (McKibben actuators) contract when pressurized and provide high force-to-weight ratios with inherent compliance. Dielectric elastomer actuators (DEAs) deform in response to electric fields, offering high strain and energy density at the cost of high driving voltages (typically 1-10 kV).
Shape memory alloy (SMA) actuators change shape when heated above their transformation temperature. Nitinol (nickel-titanium) SMA wires contract up to 8% when electrically heated and return to their original length on cooling. SMA actuators are used in micro-robotics and soft grippers where simplicity and size constraints outweigh low efficiency and limited bandwidth.
Electrostatic actuators use Coulomb forces between charged electrodes to produce motion. MEMS electrostatic actuators are widespread in micro-robotics, while larger-scale electrostatic clutches and brakes are emerging in haptic feedback and exoskeleton applications.
Sensor Fusion Techniques
Individual sensors have complementary strengths and weaknesses. Sensor fusion combines multiple sensors to produce a more accurate and reliable estimate than any single sensor alone. The complementary filter fuses accelerometer and gyroscope data for attitude estimation — gyroscope drift at low frequencies is corrected by accelerometer gravity vector measurement at high frequencies. Complementary filters are computationally trivial and widely used in drone flight controllers.
The extended Kalman filter (EKF) provides optimal state estimation for nonlinear systems. The prediction step propagates the state estimate forward using a motion model. The update step corrects the prediction using sensor measurements weighted by their uncertainty. The EKF maintains a covariance matrix representing estimate uncertainty, which naturally weights each sensor by its confidence. Factor graphs provide a modern alternative to EKFs for smoothing-based estimation, estimating the entire trajectory from all measurements rather than filtering sequentially. GTSAM and g2o libraries implement factor graph optimization for SLAM and sensor fusion.
Actuator Control
Pulse-width modulation (PWM) is the standard method for controlling DC motor voltage. The PWM signal switches a MOSFET at high frequency (20-50 kHz) with a duty cycle controlling average voltage. H-bridge circuits enable bidirectional motor control. Position control implements a cascaded control architecture: the outer position loop runs at 10-100 Hz computing velocity setpoint, the velocity loop runs at 100-1000 Hz computing torque setpoint, and the inner current loop runs at 10-20 kHz computing PWM duty cycle.
Current sensing enables torque control. Sense resistors in the motor drive circuit measure phase currents with microsecond response. Torque control prevents motor overload and enables compliant interaction with the environment. Field-oriented control (FOC) for BLDC motors requires fast ADC sampling of phase currents and a dedicated control loop running at 10-20 kHz. The SimpleFOC library provides open-source implementations for Arduino and STM32 platforms.
FAQ
What LiDAR should I buy for my first robot?
The RPLidar A2M12 ($250) or A3M1 ($350) provides adequate 2D scanning for SLAM and navigation on a research or educational robot. For 3D, the Ouster OS0-64 provides high-quality 3D data at approximately $3,000.
What is the difference between a servo motor and a DC motor?
A servo motor includes integral position feedback and control electronics, accepting position commands directly. A DC motor requires external encoder, controller, and drive electronics to achieve position control. Servos simplify integration; DC motors offer higher performance and customization.
How do I choose between I2C and SPI for sensor communication?
I2C uses fewer pins (2 wires vs 4+ for SPI) and supports multiple devices on a shared bus. SPI offers higher data rates (10+ MHz vs 400 kHz for I2C) and simpler protocol implementation. Use I2C for low-bandwidth sensors (IMU at 100 Hz) and SPI for high-bandwidth sensors (cameras, high-speed ADCs).
What sensor fusion algorithm should I use?
Use Madgwick or Mahony complementary filters for simple IMU attitude estimation on microcontrollers. Use the extended Kalman filter for full state estimation with multiple sensor types. Use factor graph optimization for offline batch estimation or SLAM applications.
Can I use camera data to replace LiDAR?
Visual SLAM and depth estimation have improved dramatically, but cameras do not directly measure distance — depth must be inferred. For outdoor, well-lit environments with sufficient texture, stereo vision or monocular depth estimation can partially replace LiDAR. For indoor, low-light, or safety-critical applications, LiDAR remains necessary.
Related: Study embedded robotics for driver development and real-time interfacing. Learn robot kinematics for understanding how sensor data relates to robot geometry. Explore AI in robotics for perception algorithms that process sensor data.