Robot Anatomy — Sensors and Actuators
A robot's body is built from two complementary categories of hardware: sensors that gather information about the world and actuators that act on it. Between them sits the compute platform that runs the software doing the sensing and planning. Understanding this hardware stack is essential context for everything that follows in this course.
Proprioceptive vs Exteroceptive Sensors
Sensors are divided into two categories based on what they measure.
Proprioceptive sensors measure the robot's own internal state: the position of its joints, its speed, its orientation in space, the forces being applied at its end effector. The word comes from the Latin for "one's own." A robot's proprioceptive system is its awareness of itself.
Exteroceptive sensors measure the external environment: where objects are, what the terrain looks like, how far away obstacles are, what the ambient temperature is. These sensors give the robot its awareness of the world around it.
Most real robots use both. A robotic arm needs proprioceptive joint encoders to know where its own links are and exteroceptive cameras or force sensors to know how it is interacting with the objects it is handling.
Key Proprioceptive Sensors
Encoders are the most fundamental proprioceptive sensor. An encoder is attached to a motor shaft and counts rotations, giving the software an accurate measure of how far a joint has moved. Incremental encoders count rotation from a reference point. Absolute encoders report the exact angular position at any time, even after power-off. Encoders are what allow a robotic arm to move to a precise position and know it has arrived.
Inertial Measurement Units (IMUs) combine accelerometers and gyroscopes into a single sensor package. Accelerometers measure linear acceleration along each axis; gyroscopes measure angular velocity. Integrating these measurements over time gives the robot's velocity, orientation and (with further integration) position. IMUs are critical for mobile robots and drones where continuous motion must be tracked. Their weakness is drift: small measurement errors accumulate over time until the position estimate becomes unreliable. This is why IMUs are almost always combined with other sensors.
Force and torque sensors are typically mounted at a robot's wrist, measuring the forces and torques being applied between the robot's end effector and the objects it is interacting with. This feedback is essential for tasks that require delicacy: assembly, surgery or any contact with humans. Without force sensing, a robot has no way to know if it is pushing too hard or not hard enough.
Key Exteroceptive Sensors
Cameras are the richest source of environmental information. A standard RGB camera provides 2D images. Stereo cameras (two lenses with a known separation) enable depth estimation from disparity (the same principle as human binocular vision). Cameras provide high-resolution texture and colour information that other sensors cannot, making them essential for object recognition, reading labels and understanding fine scene detail. Their weaknesses are sensitivity to lighting conditions, the difficulty of extracting 3D geometry from 2D images and high data rates.
LIDAR (Light Detection And Ranging) emits laser pulses and measures the time for each pulse to bounce back, producing accurate distance measurements. A spinning LIDAR produces a full 360-degree point cloud of the environment: a 3D map of where surfaces are, accurate to centimetres, many times per second. LIDAR is the primary long-range sensor in autonomous vehicles and mobile robots. Its weaknesses are cost (though prices have fallen dramatically), inability to work well in rain or fog and the lack of texture or colour information.
Depth cameras combine a standard camera with a depth measurement mechanism: structured light (projecting a known infrared pattern and measuring its distortion), stereo (two cameras) or time-of-flight (measuring the return time of infrared light). Intel RealSense and Microsoft Kinect are well-known examples. Depth cameras are cheaper and lighter than LIDAR, work well at short range and produce colour plus depth simultaneously. They are common on indoor robots and manipulators.
Ultrasonic sensors emit a sound pulse and measure the echo delay. Simple, cheap and effective for close-range obstacle detection. Used in robot vacuum cleaners, parking sensors on cars and proximity detection. Low resolution: they cannot distinguish fine object shapes.
Actuators
If sensors are a robot's nervous system receiving signals, actuators are its muscles executing commands.
Servo motors (DC and brushless) are the dominant actuator in most robots. A servo motor is an electric motor combined with a position sensor and a controller that maintains a commanded position. They are precise, responsive and easy to control digitally. The robotic arms used in industrial manufacturing, collaborative robots and humanoids all run primarily on electric motors. Gearboxes are often added to trade speed for torque.
Hydraulic actuators use pressurised fluid to generate force. They produce far greater force than equivalently sized electric motors, making them the actuator of choice for heavy industrial applications and high-performance legged robots. Boston Dynamics' early Atlas was hydraulically powered. Hydraulics' downsides are weight, complexity, the need for a pump and reservoir and the risk of fluid leaks.
Pneumatic actuators use compressed air. They are fast, lightweight and inexpensive. Common in industrial grippers, pick-and-place systems and soft robotics (where compliant silicone structures are inflated to create gentle, flexible grippers). Their limitation is that compressed air supply must be continuously available, making them impractical for untethered mobile robots.
The Hardware-Software Interface
The connection between software and hardware is one of the most important and underappreciated aspects of robotics.
Software sends commands to actuators and receives data from sensors through a combination of low-level device drivers, real-time controllers running on dedicated microcontrollers and higher-level software frameworks like ROS (which you will meet in the next lesson). A robotic arm may have a controller running at 1,000 Hz to maintain stable joint position while the high-level planning software makes decisions at 10 Hz.
Calibration is the process of precisely measuring sensor characteristics and geometric relationships so that software can correctly interpret raw sensor data. A camera must be calibrated so that a pixel at a known location in the image corresponds to a known point in 3D space. A force sensor must be calibrated so that its raw voltage readings map correctly to Newtons. Without calibration, the mathematics of perception does not hold.
Latency, the delay between sensing something and acting on it, is a critical concern. High latency means the robot is always acting on stale information. In fast dynamic tasks, even tens of milliseconds of latency can cause instability. Minimising latency through careful hardware design and software architecture is a recurring engineering challenge.
Quiz: What is the difference between a proprioceptive and an exteroceptive sensor? Give one example of each. What are the main trade-offs between LIDAR and depth cameras as exteroceptive sensors?