How integrated circuits power robots

Robotic systems combine sensing and processing with mechanisms that create movement. Vision systems interpret images, software can estimate location, and actuators generate physical work. (Cognex: machine vision) (MathWorks: SLAM) (University of Rochester: actuators)

1. SenseCollect measurements
2. ComputeEstimate and decide
3. ActCommand physical motion
4. FeedbackMeasure what happened

Use this loop to organize the terminology. A sensor supplies measurements, a processor executes calculations, a driver provides electrical drive, and an actuator produces mechanical work. (Texas Instruments: motor control) (University of Rochester: actuators) Software names such as SLAM and ROS 2 belong to another layer of that same system. (MathWorks: SLAM) (ROS 2 documentation)

A depth camera represents distances within a scene; LiDAR uses light-based ranging to produce spatial measurements; an IMU measures acceleration and angular motion using inertial sensors. (RealSense: depth cameras) (MathWorks: SLAM) (Analog Devices: IMU documentation)

Machine vision uses images to inspect, locate, or identify objects. Sensor fusion combines information from different sensors, while SLAM builds a map and estimates location within it. (Cognex: machine vision) (MathWorks: SLAM)

These are related functions, not interchangeable labels. A camera does not become a complete navigation system simply because it provides depth, and a SLAM algorithm still depends on its inputs and implementation.

CPUs support general program execution, GPUs accelerate suitable parallel workloads, and NPUs specialize in neural-network operations. (Intel: CPU vs. GPU) (Intel: AI accelerators) An integrated SoC can bring several processing and interface functions together. (Arm: SoC development)

Edge AI means the AI application runs near the data source. A robot can use a trained model locally for inference, while model training and updates are handled elsewhere. (NVIDIA: edge AI)

ROS 2 is a collection of libraries and tools for robot applications. An RTOS is an operating system designed for predictable responses within timing constraints; neither term names a semiconductor chip. (ROS 2 documentation) (FreeRTOS: RTOS fundamentals)

When reading a system diagram, label hardware, software, algorithms, and physical mechanisms separately. That simple exercise prevents an “AI processor” description from hiding the rest of the robot’s design.

For more detail on the computing layer, see GPU vs. NPU vs. CPU. For memory and packaging terminology, use the HBM and chiplet definitions.

A servo controller uses feedback to move toward a desired position or velocity. A motor driver supplies electrical drive, and an actuator converts energy into physical work. (University of Rochester: actuators) (Texas Instruments: motor control)

An encoder can provide position or motion feedback. Wheel measurements can also feed odometry, which estimates a vehicle’s position and orientation from movement information. (Texas Instruments: motor control) (MathWorks: wheel odometry)

Real-time control concerns meeting defined timing requirements, not just achieving a high average processing rate. RTOS task scheduling can help manage execution, but cannot make impossible deadlines feasible. (FreeRTOS: RTOS fundamentals)

Communication is another layer. CAN provides message-based communication between nodes; EtherCAT is a real-time industrial Ethernet technology used in automation. (Texas Instruments: CAN fundamentals) (EtherCAT Technology Group)