AMR perception has progressed to neural processing and inference on the robot. MEMS inertial sensors with embedded machine learning cores classify events and control architectures distribute inference workloads across MCU, MPU, and NPU by task, instead of concentrating in one central processor. The design decision is now where each decision is made: in the sensor, at the control loop, or in central compute.
Main functions
The following table shows the role of various sensing and control elements in an AMR.
| Component | Category | Role in the AMR |
|---|---|---|
| Vision sensing (camera, ToF, LiDAR) | Sensing | Perceive the surroundings for obstacle detection and mapping |
| Motion sensing (IMU) | Sensing | Track the robot's own movement for odometry and stability |
| Environmental sensing (temperature) | Sensing | Monitor operating conditions, feed thermal protection |
| MCU | Processing | Handle real-time, deterministic control tasks |
| MPU | Processing | Run navigation, mapping, and path planning |
| NPU | Processing | Accelerate AI perception workloads such as object detection |
| EEPROM | Memory | Store calibration and configuration across power cycles |
| Secure element | Security | Authenticate the robot to the fleet, verify firmware integrity |