Sensors and Actuators: The Interface Between Digital and Physical

Purpose

This chapter examines the fundamental hardware components that enable Physical AI: sensors that perceive the world and actuators that act upon it. Understanding these components is essential for designing robust embodied systems.

Sensors: Perceiving the Physical World

Sensors convert physical phenomena into digital signals. Robotics sensors are categorized by what they measure:

1. Proprioceptive Sensors (Internal State)

Proprioception: sensing the robot's own configuration and motion.

Joint Encoders

Function: Measure joint angles or positions.

Types:

• Absolute Encoders: Provide unique position code for each angle
• Incremental Encoders: Count rotations from reference position
• Potentiometers: Analog voltage proportional to angle

Specifications:

• Resolution: 12-bit (4096 positions/revolution) to 18-bit (262,144 positions)
• Accuracy: ±0.1° to ±0.01°
• Update Rate: 1-10 kHz

Application: Every robot joint has an encoder for closed-loop control.

Limitation: Encoders measure joint angle, not link position (requires forward kinematics).

Inertial Measurement Unit (IMU)

Function: Measure acceleration and angular velocity.

Components:

• 3-axis Accelerometer: Linear acceleration (gravity + motion)
• 3-axis Gyroscope: Angular velocity (rotation rate)
• 3-axis Magnetometer (optional): Magnetic field (compass heading)

Output: 9 DOF (degrees of freedom) sensor fusion → orientation estimate.

Specifications:

• Accelerometer Range: ±2g to ±16g
• Gyroscope Range: ±250°/s to ±2000°/s
• Update Rate: 100-1000 Hz
• Noise: Gyro drift (0.01-1 °/s), accelerometer noise (0.001-0.01 g)

Application: Humanoid balance control, drone stabilization, vehicle localization.

Challenge: Drift accumulates over time (requires periodic correction from other sensors).

Force/Torque Sensors

Function: Measure contact forces and torques.

Types:

• Strain Gauge Based: Deformation → resistance change → force
• Capacitive: Force → gap change → capacitance change
• Optical: Force → light intensity change

Placement:

• Wrist: 6-axis force/torque for manipulation
• Ankle/Foot: Ground reaction forces for balance
• Joints: Torque sensing for compliance control

Specifications:

• Range: 10N-10kN (force), 1Nm-1kNm (torque)
• Resolution: 0.1N-10N
• Update Rate: 1-10 kHz

Application: Compliant manipulation, contact detection, grasp force regulation.

Example: A robot adjusting grip force when picking up a fragile egg vs. a heavy wrench.

2. Exteroceptive Sensors (External Environment)

Exteroception: sensing the surrounding environment.

Cameras

Function: Capture visual information as 2D images.

Types:

RGB Cameras:

• Resolution: 640×480 (VGA) to 3840×2160 (4K)
• Frame Rate: 30-120 FPS
• Field of View (FOV): 60-180°
• Use Case: Object recognition, visual servoing, semantic understanding

Stereo Cameras:

• Two cameras with known baseline → depth via triangulation
• Depth Range: 0.5m-20m
• Depth Accuracy: 1-5% of distance
• Use Case: 3D reconstruction, obstacle avoidance

Event Cameras (Neuromorphic):

• Pixel-level change detection (not full frames)
• Latency: under 1ms
• Dynamic Range: 120dB (vs. 60dB for traditional)
• Use Case: High-speed motion tracking, low-light conditions

Specifications:

• Dynamic Range: 60-120 dB (ability to handle bright/dark simultaneously)
• Latency: 10-100ms (traditional), under 1ms (event cameras)
• Power: 1-5W

Challenge: Lighting variation, motion blur, occlusion.

Depth Sensors

Function: Directly measure distance to objects.

Types:

LiDAR (Light Detection and Ranging):

• Principle: Time-of-flight of laser pulse
• Range: 0.1m-100m
• Accuracy: 1-5cm
• Scan Rate: 5-40 Hz (rotating) or 10-100 Hz (solid-state)
• Points/Second: 100k-10M
• Use Case: Autonomous vehicles, mapping, obstacle detection

Structured Light (e.g., Kinect):

• Project pattern → measure distortion → compute depth
• Range: 0.5m-4.5m
• Resolution: 640×480
• Frame Rate: 30 FPS
• Use Case: Indoor robotics, gesture recognition

Time-of-Flight (ToF) Cameras:

• Measure time for modulated light to return
• Range: 0.1m-10m
• Resolution: 320×240
• Frame Rate: 30-60 FPS
• Use Case: Short-range obstacle detection, hand tracking

Tradeoffs:

• LiDAR: Long range, high accuracy, expensive ($500-$10k)
• Structured Light: Short range, sensitive to lighting, affordable (under $100)
• ToF: Medium range, compact, moderate cost ($200-$500)

RADAR (Radio Detection and Ranging)

Function: Detect objects using radio waves.

Specifications:

• Range: 1m-300m
• Accuracy: 10cm-1m
• Update Rate: 10-100 Hz
• Weather Resistance: Excellent (works in rain, fog, dust)

Use Case: Automotive (adaptive cruise control), long-range obstacle detection.

Limitation: Low resolution (cannot distinguish small objects), limited elevation information.

Ultrasonic Sensors

Function: Measure distance using sound waves.

Specifications:

• Range: 2cm-4m
• Accuracy: 1cm
• Beam Angle: 15-30° (wide cone)
• Update Rate: 10-50 Hz

Use Case: Parking sensors, close-range obstacle detection, cliff detection.

Limitation: Specular reflection (smooth surfaces deflect signal), slow speed of sound.

3. Tactile Sensors (Touch)

Tactile sensing provides contact information crucial for manipulation.

Pressure Sensors

Types:

• Resistive: Force → resistance change
• Capacitive: Force → capacitance change
• Piezoelectric: Force → voltage

Array Configurations:

• Fingertip Arrays: 4×4 to 16×16 taxels (tactile pixels)
• Palm Arrays: Lower density (8×8)
• Whole-Body Skin: Distributed pressure sensing

Specifications:

• Pressure Range: 0.1N-100N
• Spatial Resolution: 1-10mm
• Update Rate: 100-1000 Hz

Use Case: Grasp stability detection, contact force regulation, slip detection.

Example: A robot detecting object slipping and increasing grip force before dropping.

Force-Sensing Resistors (FSR)

Principle: Pressure → conductive paths → resistance decrease.

Advantages: Thin (0.5mm), flexible, low cost (under $5).

Disadvantages: Nonlinear response, hysteresis, drift over time.

Use Case: Simple touch detection (presence/absence), low-cost prototyping.

4. Multi-Modal Sensor Fusion

Real robots combine multiple sensor types for robustness:

Example: Humanoid Perception Stack:

• Vision: RGB-D cameras for object recognition
• Depth: LiDAR for environment mapping
• Proprioception: IMU + joint encoders for state estimation
• Tactile: Pressure arrays in hands for manipulation

Fusion Algorithm: Kalman Filter, Particle Filter, or learned sensor fusion (neural networks).

Benefits:

• Redundancy: Sensor failure doesn't cause complete system failure
• Complementary Information: Vision provides semantics, LiDAR provides geometry
• Improved Accuracy: Fusing noisy sensors reduces uncertainty

Challenge: Different update rates, coordinate frames, and noise characteristics.

Actuators: Acting on the Physical World

Actuators convert electrical energy into mechanical motion.

1. Electric Motors

Most common in modern robotics.

Brushless DC (BLDC) Motors

Principle: Rotating magnetic field drives rotor.

Advantages:

• High efficiency (85-95%)
• Long lifespan (no brush wear)
• High power density (W/kg)
• Precise control (via encoder feedback)

Disadvantages:

• Requires electronic speed controller (ESC)
• Higher cost than brushed motors

Specifications:

• Torque: 0.01Nm-10Nm (direct drive)
• Speed: 1000-50,000 RPM (no-load)
• Power: 10W-5kW (typical robotics range)

Use Case: Drones (high speed), robotic arms (moderate torque).

Servo Motors

Definition: Motor + gearbox + controller + encoder in single package.

Types:

• Hobby Servos: 180° range, PWM control, under 10Nm
• Industrial Servos: Multi-turn, field bus communication, >100Nm

Advantages: Closed-loop position control out-of-the-box.

Use Case: Hobby robotics, joints with limited range.

Stepper Motors

Principle: Discrete step rotation (1.8° or 0.9° per step typical).

Advantages:

• Open-loop control (no encoder needed)
• Holding torque at rest
• Precise positioning (if no missed steps)

Disadvantages:

• Can lose steps under excessive load (no feedback)
• Lower efficiency than BLDC
• Resonance at certain speeds

Use Case: 3D printers, CNC machines, precise positioning tasks.

2. Gearboxes and Transmissions

Motors produce high speed, low torque. Gearboxes trade speed for torque.

Harmonic Drives (Strain Wave Gears)

Principle: Flexible spline deforms inside circular spline → high ratio in compact form.

Gear Ratios: 50:1 to 160:1 (single stage).

Advantages:

• Zero backlash (critical for precision)
• Compact, coaxial design
• High torque capacity

Disadvantages:

• Expensive ($500-$5000)
• Friction and hysteresis

Use Case: Robot arms, humanoid joints (shoulder, hip).

Planetary Gearboxes

Principle: Central sun gear, planet gears, ring gear → multiple gear stages.

Gear Ratios: 3:1 to 100:1.

Advantages:

• Load distribution across multiple gears (high torque)
• Efficient (90-95%)

Disadvantages:

• Backlash (small gap between teeth)

Use Case: Heavy-duty industrial robots.

3. Series Elastic Actuators (SEA)

Principle: Spring in series between motor and load.

Advantages:

• Compliance: Spring absorbs shocks (safe human interaction)
• Force Sensing: Spring deflection → force measurement (no separate sensor)
• Energy Storage: Spring stores energy (walking efficiency)

Disadvantages:

• Reduced bandwidth (spring introduces lag)
• Larger, heavier than rigid actuators

Use Case: Humanoids requiring safe human contact (ASIMO, Cassie), prosthetics.

Example: If robot bumps into person, spring compresses (limiting impact force) instead of rigid collision.

4. Hydraulic Actuators

Principle: Pressurized fluid drives piston → linear motion.

Advantages:

• High Power-to-Weight Ratio: 10× better than electric
• High Force: Can lift hundreds of kg
• Smooth Motion: Fluid damping

Disadvantages:

• Complexity (pump, valves, reservoir, hoses)
• Leakage risk (fluid maintenance)
• Noisy (pump)
• Difficult to control precisely

Use Case: Boston Dynamics Atlas (explosive power for jumping), construction equipment.

Specifications:

• Pressure: 100-350 bar
• Force: 1kN-100kN
• Speed: 0.1-1 m/s

5. Pneumatic Actuators

Principle: Compressed air drives piston or inflates chamber.

Types:

• Piston Cylinders: Linear motion
• Soft Pneumatic Actuators: Flexible chambers (soft robotics)

Advantages:

• Inherently compliant (safe interaction)
• Simple, lightweight
• Naturally backdrivable

Disadvantages:

• Low precision (air compressibility)
• Requires air compressor
• Low force compared to hydraulics

Use Case: Soft grippers, human-safe manipulation, pneumatic artificial muscles.

Example: Soft robotic gripper gently grasping delicate fruit without bruising.

6. Novel Actuator Technologies

Artificial Muscles

Types:

• Pneumatic Artificial Muscles (PAM): Braided sleeve contracts when inflated
• Shape Memory Alloy (SMA): Wire contracts when heated
• Electroactive Polymers (EAP): Polymer changes shape under voltage

Potential: Biomimetic motion, high power density, silent operation.

Challenge: Control difficulty, slow response (SMA), low force (EAP).

Sensor-Actuator Integration

Effective Physical AI requires tight sensor-actuator loops:

Control Loop Example: Joint Position Control

1. Sense: Encoder measures current joint angle θ_current
2. Compute Error: e = θ_desired - θ_current
3. Control Law: Motor voltage V = Kp × e + Kd × (de/dt) (PID control)
4. Actuate: Motor rotates joint
5. Repeat: 100-1000 Hz

Multi-Modal Feedback

Visual Servoing:

• Input: Camera image of target object
• Process: Compute image error (desired - actual pixel coordinates)
• Output: Velocity command to robot arm
• Feedback: Camera continuously updates target position

Force Control:

• Input: Force sensor reading F_current
• Process: Compute force error e_F = F_desired - F_current
• Output: Position adjustment to increase/decrease contact force
• Feedback: Force sensor continuously updates

Key Specifications and Selection Criteria

For Sensors

Criterion	Consideration
Range	Minimum and maximum measurable values
Resolution	Smallest detectable change
Accuracy	Error vs. true value
Precision	Repeatability of measurements
Update Rate	Measurements per second (Hz)
Latency	Delay from phenomenon to digital signal
Power	Energy consumption
Cost	Per-unit price
Size/Weight	Physical constraints

For Actuators

Criterion	Consideration
Torque/Force	Maximum output
Speed	Maximum rotational/linear velocity
Power	Continuous and peak ratings
Efficiency	Energy conversion ratio
Bandwidth	Frequency response (control speed)
Backlash	Dead zone in gear systems
Compliance	Rigidity vs. flexibility
Cost	Per-unit price
Size/Weight	Physical constraints

Practical Examples

Example 1: Robotic Arm Joint

Requirements: Lift 5kg at 0.5m reach, position accuracy 1mm.

Sensor Selection:

• Joint Encoder: 14-bit absolute encoder (0.02° resolution)
• Force/Torque Sensor: 6-axis wrist sensor (100N range, 0.1N resolution)

Actuator Selection:

• Motor: 200W BLDC motor (3Nm continuous torque)
• Gearbox: 50:1 harmonic drive (zero backlash)
• Total Torque: 3Nm × 50 = 150Nm (sufficient for 5kg × 0.5m × 9.8 = 24.5Nm)

Control: Cascaded PID (position loop at 1kHz, torque loop at 10kHz).

Example 2: Humanoid Foot Pressure Sensing

Requirements: Detect contact forces for balance control, 100 Hz update.

Sensor Selection:

• Foot Array: 8×8 FSR array (64 taxels)
• Placement: Four force sensors at heel, toe, and both sides of foot
• Processing: Sum all sensor readings → total ground reaction force

Use: Compute Center of Pressure (CoP) → balance controller adjusts joint torques to maintain CoP within support polygon.

Key Takeaways

1. Sensors convert physical phenomena into digital signals, categorized as proprioceptive (internal state), exteroceptive (environment), and tactile (contact).

2. Essential robot sensors include joint encoders (position), IMUs (orientation), cameras (vision), depth sensors (LiDAR/stereo), and force/torque sensors (contact).

3. Sensor fusion combines multiple modalities for robustness, redundancy, and complementary information.

4. Actuators convert electrical energy to mechanical motion, with electric motors (BLDC, servo, stepper) dominating modern robotics.

5. Gearboxes trade speed for torque, with harmonic drives preferred for precision (zero backlash) and planetary gears for heavy loads.

6. Specialized actuators include Series Elastic Actuators (compliance, force sensing), hydraulics (high power), and pneumatics (soft interaction).

7. Sensor-actuator integration requires real-time control loops (100-1000 Hz) with tight feedback for precision and stability.

8. Selection criteria balance range, resolution, accuracy, update rate, power, cost, and physical constraints based on application requirements.

---

Next Chapter: System architecture—how sensors, actuators, compute, and AI components integrate into cohesive Physical AI systems.