Sensors

Robots need to sense themselves and their world. Learn encoders and resolvers, force/torque sensors, IMUs, proximity and vision sensors, LiDAR and time-of-flight, resolution vs accuracy, and sensor fusion.

A robot is blind and numb without sensors. Internal (proprioceptive) sensors measure its own joints; external (exteroceptive) sensors measure the environment — parts, people and obstacles.

Choosing sensors means understanding resolution, accuracy, range, speed, noise and cost.

An analogy

Encoders are the robot's sense of where its limbs are (like your proprioception); cameras and LiDAR are its eyes; force sensors are its sense of touch.

Proprioceptive sensors

Incremental encoders count pulses (A/B quadrature) from a reference; absolute encoders know position at power-up (multi-turn with battery or gear). Resolvers are rugged analogue alternatives. Motor current gives an estimate of torque.

Exteroceptive sensors

Proximity (inductive, capacitive, photoelectric), 2D/3D vision (structured light, stereo, ToF cameras), LiDAR (laser scanners for mapping and safety), ultrasonic ranging, 6-axis force/torque sensors at the wrist for assembly and polishing, and tactile skins.

Resolution, accuracy and repeatability

Resolution is the smallest change you can detect; accuracy is closeness to the truth; repeatability is consistency. An encoder may resolve 0.0004° but link deflection, gear error and thermal growth limit tool accuracy.

Sensor fusion

IMUs (accelerometer + gyro) drift; GPS/vision/LiDAR are slower but absolute. Kalman and complementary filters fuse them for state estimation; SLAM builds maps while localising. Hand–eye calibration relates camera frames to robot frames for vision-guided picking.

Formulas

• Quadrature resolution: CPR = 4 × PPR, Δθ = 360°/CPR
• Joint-side resolution: Δθ_joint = 360°/(CPR × N)
• Tool resolution: Δs = R · Δθ (rad)
• Time of flight: d = c · t_f / 2

Worked example: Encoder resolution through a gearbox

Given: 2500 PPR quadrature encoder on the motor; Gear ratio 100; Tool 0.7 m from the joint

Find: Joint and tool resolution.

• CPR = 10 000 → joint counts = 10⁶ per rev
• Δθ = 360°/10⁶ = 0.00036°
• Δs = 0.7 × 2π/10⁶ = 4.4 µm

Answer: 0.00036° at the joint, 4.4 µm at the tool

Worked example: LiDAR range

Given: Round-trip time 66.7 ns

Find: Distance.

• d = 3 × 10⁸ × 66.7 × 10⁻⁹/2 = 10.0 m

Answer: 10 m

Common mistakes

• Wrong: Confusing resolution with accuracy. — Right: High encoder resolution doesn't guarantee accurate tool position.
• Wrong: Skipping camera-to-robot calibration. — Right: Perform hand–eye calibration and verify with test points.

Applications

• Bin picking: 3D vision locates randomly oriented parts for the robot to grasp.
• Assembly & finishing: Force control for peg-in-hole insertion, grinding and polishing.
• Safety: Safety laser scanners and light curtains slow or stop robots when people approach.

Key takeaways

• Encoders for joints, vision/LiDAR for the world, F/T for touch.
• Resolution ≠ accuracy ≠ repeatability.
• d = ct/2 for time-of-flight sensors.
• Fuse sensors and calibrate frames.

Going further

• Event cameras: Asynchronous pixels with microsecond latency for high-speed tracking.
• Tactile sensing: Optical tactile sensors (e.g. GelSight-type) sense contact geometry for dexterous manipulation.
• Functional safety: Safety-rated sensors (PL d/e, SIL 2/3) are required for human–robot collaboration.