Robot Sensors: Cameras, LiDAR and Encoders

Sensor families

• /robot-sensors — Vision sensors give robots color, shape, motion, texture and heat cues. They usually sit on the head, mast, wrist, gripper, mobile base or inspection payload. The hard part is not only camera selection. It is lighting, calibration, lens choice, synchronization and software that can tell uncertainty from a real object.
• /robot-sensors — LiDAR measures distance with laser scanning. A 2D LiDAR can be enough for indoor AMR navigation, while 3D LiDAR is useful for uneven terrain, mapping, outdoor perception and inspection. Safety LiDAR is different because it is certified for protective fields and machine safety.
• /robot-sensors — Force, torque and tactile sensors tell the robot what contact feels like. They matter when a robot must insert a part, sand a surface, close a gripper gently, recover from slip or learn contact rich manipulation. This is one of the hardest sensor families because the signal depends on mechanics, calibration and control loops.
• /robot-sensors — IMUs measure acceleration and angular velocity. Some units include magnetometers, barometers or onboard fusion. Drones, humanoids and quadrupeds depend on motion sensing, but IMU data drifts and must be fused with encoders, cameras, LiDAR, GNSS or contact sensors.
• /robot-sensors — Proximity sensors are simple, cheap and often more reliable than complex perception for one narrow job. They can tell a gripper that a part is present, a robot cell that a fixture is loaded or a small robot that an obstacle is close. Their weakness is that every technology has blind spots.
• /robot-sensors — Encoders tell the robot where a shaft, wheel or joint is. A robot arm without good position feedback cannot hold accurate paths. A mobile base without encoder feedback cannot estimate wheel motion. Encoders look simple, but mounting, alignment, cable noise and absolute reference matter more than many beginners expect.
• /robot-sensors — Safety sensors are not ordinary perception sensors. They protect people and machines under defined standards, wiring and validated logic. A vision model that detects humans is not a replacement for a certified safety circuit in an industrial cell.
• /robot-sensors — Tactile skin is still an emerging field. It can provide local pressure, shear, contact location or high resolution surface geometry. The promise is large for humanoid hands and Physical AI, but integration is hard because robot skin needs wiring, protection, calibration, replacement parts and software that can use dense touch data.

Vision sensors give robots color, shape, motion, texture and heat cues. They usually sit on the head, mast, wrist, gripper, mobile base or inspection payload. The hard part is not only camera selection. It is lighting, calibration, lens choice, synchronization and software that can tell uncertainty from a real object.

• RGB cameras: A warehouse robot can use RGB vision to read package markings while another depth sensor estimates shape.
• Stereo cameras: A delivery robot can estimate curb geometry and obstacle distance with a stereo baseline and calibrated lenses.
• Depth cameras: A robot arm can combine depth and RGB to segment an object before a gripper closes.
• Global shutter cameras: A high speed inspection robot can avoid distorted images when tracking moving parts on a conveyor.
• Event cameras: A research robot can track fast moving objects where a normal camera blurs or misses the event.
• Thermal cameras: A plant inspection robot can detect overheating motors or electrical cabinets without touching them.

LiDAR measures distance with laser scanning. A 2D LiDAR can be enough for indoor AMR navigation, while 3D LiDAR is useful for uneven terrain, mapping, outdoor perception and inspection. Safety LiDAR is different because it is certified for protective fields and machine safety.

• 2D LiDAR: An AMR can use a 2D LiDAR for SLAM and obstacle avoidance in a warehouse aisle.
• 3D LiDAR: A quadruped inspection robot can map stairs, pipes and platforms with 3D point clouds.
• Safety LiDAR: A mobile platform can reduce speed in a warning zone and stop in a protective zone near workers.
• Mapping LiDAR: A facility mapping robot can combine LiDAR, IMU and wheel odometry to build a map before fleet deployment.

Force, torque and tactile sensors tell the robot what contact feels like. They matter when a robot must insert a part, sand a surface, close a gripper gently, recover from slip or learn contact rich manipulation. This is one of the hardest sensor families because the signal depends on mechanics, calibration and control loops.

• Wrist force torque sensors: A cobot can keep a near constant contact force while polishing a curved part.
• Joint torque sensing: A legged robot can adjust foot contact when torque feedback shows unexpected ground reaction.
• Tactile fingertip sensors: A robot hand can detect that a smooth object is slipping before it falls.
• Gripper force sensing: A packaging robot can grip soft items with lower force when contact is detected early.

IMUs measure acceleration and angular velocity. Some units include magnetometers, barometers or onboard fusion. Drones, humanoids and quadrupeds depend on motion sensing, but IMU data drifts and must be fused with encoders, cameras, LiDAR, GNSS or contact sensors.

• Accelerometers: A humanoid controller can detect a body acceleration spike during a stumble.
• Gyroscopes: A drone can stabilize attitude using high rate gyro data between camera frames.
• Magnetometers: A small field robot may use magnetometer heading when GNSS and visual landmarks are weak.
• Industrial IMU and AHRS modules: A mapping robot can fuse IMU with LiDAR for smoother localization through vibration.

Proximity sensors are simple, cheap and often more reliable than complex perception for one narrow job. They can tell a gripper that a part is present, a robot cell that a fixture is loaded or a small robot that an obstacle is close. Their weakness is that every technology has blind spots.

• Time of flight sensors: A small mobile robot can detect a nearby wall before the bumper touches it.
• Ultrasonic sensors: A service robot can use ultrasonic modules as slow speed backup obstacle detection.
• Infrared proximity sensors: A line robot can detect whether a small part is present before a pick attempt.
• Capacitive and inductive sensors: An industrial robot cell can verify that a metal workpiece is seated before a welding cycle.

Encoders tell the robot where a shaft, wheel or joint is. A robot arm without good position feedback cannot hold accurate paths. A mobile base without encoder feedback cannot estimate wheel motion. Encoders look simple, but mounting, alignment, cable noise and absolute reference matter more than many beginners expect.

• Absolute encoders: A robot arm joint can know its angle immediately after the controller boots.
• Incremental encoders: A differential drive robot can estimate wheel motion with two incremental encoders.
• Magnetic encoders: A compact gripper can use magnetic angle feedback where optical alignment is difficult.
• Optical encoders: A precision robot arm axis can use optical feedback for accurate path tracking.

Safety sensors are not ordinary perception sensors. They protect people and machines under defined standards, wiring and validated logic. A vision model that detects humans is not a replacement for a certified safety circuit in an industrial cell.

• Safety scanners: An AGV can switch between warning and stop zones depending on speed and aisle geometry.
• Safety light curtains: A robot cell can stop when an operator reaches through a loading opening.
• Emergency stop and enabling systems: A technician can use an enabling switch while jogging a robot at reduced speed.
• Collision detection sensors: A cobot can stop when measured joint torque exceeds an allowed contact threshold.

Tactile skin is still an emerging field. It can provide local pressure, shear, contact location or high resolution surface geometry. The promise is large for humanoid hands and Physical AI, but integration is hard because robot skin needs wiring, protection, calibration, replacement parts and software that can use dense touch data.

• Fingertip tactile sensors: A research hand can collect touch data while turning a small object inside the fingers.
• Pressure arrays: A gripper can detect whether the object is centered or only touching one edge.
• Soft robotics sensors: A soft gripper can estimate finger bending as it wraps around a fragile object.
• Humanoid hand sensing: A humanoid hand can adjust grip after detecting that the object has shifted during lift.

Buying checklist

• Define the task before the sensor: mapping, grasping, safety, inspection, balance, localization or quality control.
• Measure the real environment: sunlight, dust, water, vibration, reflective parts, transparent objects, metal frames, people and cleaning process.
• Check integration early: power, connector, cable bend radius, ROS 2 driver, SDK, time synchronization and calibration tools.
• Budget for mechanical design: brackets, lenses, protective windows, lighting, vibration isolation and spare parts often cost more than expected.
• Do not treat a perception demo as a safety system. Certified safety sensors and validated safety logic are separate engineering work.
• Plan failure detection: dirty lens, loose cable, dead pixels, encoder dropout, IMU drift, LiDAR rain noise, tactile gel wear and thermal calibration drift.

Failure modes

• Cameras
• Depth cameras
• LiDAR
• Force torque sensors
• IMUs
• Encoders
• Safety sensors
• Tactile sensors

Sensor FAQ

• What sensors are used in robots? Robots use cameras, LiDAR, depth sensors, force torque sensors, tactile sensors, IMUs, encoders, proximity sensors, safety scanners and sometimes thermal, radar, sonar or chemical sensors. The right set depends on the task and environment.
• What sensors do humanoid robots use? Humanoid robots usually combine RGB or depth cameras, IMUs, joint encoders, motor current sensing, foot or wrist force sensors and sometimes tactile hands. Public demos do not always reveal every sensor or autonomy level.
• What is the difference between LiDAR and depth cameras? LiDAR scans distance with lasers and is strong for mapping and navigation. Depth cameras estimate nearby geometry through stereo, structured light or time of flight and are common for manipulation, picking and short range perception.
• How much do robot sensors cost? Small proximity or IMU sensors can cost under $50. Depth cameras often sit in the hundreds. Industrial LiDAR, safety scanners, force torque sensors and thermal cameras can cost thousands. Prices change often and vary by region and configuration.
• What sensors are used in robot arms? Robot arms use encoders in every joint, motor current sensing, limit switches, thermal sensors, sometimes joint torque sensing and wrist force torque sensors. Vision and depth cameras are often added near the cell or wrist for picking and inspection.
• What sensors are used in AMRs? AMRs often use 2D LiDAR, safety scanners, wheel encoders, IMUs, RGB or depth cameras and sometimes 3D LiDAR. Industrial AMRs also need safety rated sensing and validated stop behavior.
• What sensors are used in grippers? Grippers can use jaw position sensing, motor current, force sensors, tactile fingertips, vacuum pressure sensors and proximity sensors. The sensor should be chosen around the object, not only the gripper model.
• What is a force torque sensor? A force torque sensor measures forces and moments, usually along six axes. Mounted at a robot wrist, it helps the robot regulate contact during insertion, sanding, assembly, polishing and compliant manipulation.
• Why do robots need encoders? Encoders tell the controller where a joint, wheel or motor shaft is. Without position feedback, a robot cannot repeat paths, control speed accurately or estimate its body state reliably.
• Which robot sensors are expensive? High end 3D LiDAR, certified safety scanners, industrial force torque sensors, precision encoders, thermal cameras and specialized tactile skins are usually expensive compared with simple IMUs or short range distance sensors.