Feature: Industrial electronics
This move to human-centric automation dramatically shifts the burden onto systems engineers
• Time-of-flight (ToF) and structured light sensors (Lidar/IR) are used for high-fidelity depth mapping and creating dense point clouds of the physical workspace. They provide precise geometry measurements, critical for collision avoidance and path planning. For industrial use, the key challenge is minimum depth resolution, often to sub-millimeter accuracy, and robust multi-sensor registration. Disparate 3D point clouds from multiple devices must be stitched together into a single, coherent map, necessitating precise temporal synchronisation and calibration to prevent spatial seams or misalignment.
• High-resolution RGB cameras capture visual data essential for object recognition and texture mapping. The critical engineering challenge here is overcoming illumination variability. Factory floors feature bright windows, shadows and harsh task lighting, requiring sophisticated image signal processing pipelines to maintain high image quality for real- time computer vision models (such as deep convolutional neural networks) that recognise diverse workpieces, tools and human safety gear.
• Inertial measurement units (IMUs) provide high-frequency, low-latency tracking of the user or robot’s position and orientation. T is is essential for stabilising the mixed reality overlay and the robot control loop. Engineers must focus on visual-inertial odometry – the process of fusing high-frequency, noisy IMU data with lower-frequency, high- accuracy visual data – to prevent sensor driſt and ensure the virtual instructions remain perfectly anchored to the physical world, a phenomenon known as preventing the MR overlay from “swimming”.
www.electronicsworld.co.uk March 2026 43
2. Edge processing unit: SLAM and semantic segmentation T e sheer volume of raw sensor data requires signifi cant on-device (edge) computation to perform initial processing. T is processing establishes the foundational intelligence of the system: • SLAM: T ese algorithms build the map and simultaneously track the sensor’s position within it. Robust SLAM implementations are paramount for industrial stability. T is demands meticulous engineering of driſt compensation (correcting cumulative errors over time) and loop closure detection (recognising a previously visited location to correct the entire map structure). T e system oſt en uses dense mapping methods over feature- based ones to ensure the high-fi delity required for precise robotic path planning.
• Real-time semantic segmentation: AI algorithms process the structured
3D data to understand the semantic meaning of the physical space. T is is the intelligence layer that distinguishes a “workpiece” from a “fi xture”, “safety zone boundary” or “human operator”. Implementing this requires high- performance deep learning models optimised for edge hardware, where effi cient computation of ground truth labelling for industrial data must be handled to run classifi cation and segmentation models in real time.
Signal processing and integrity For systems and components developers, ensuring signal integrity, electromagnetic immunity and data security in the harsh industrial environment is non-negotiable for reliable Spatial AI implementation.
Signal integrity and noise immunity T e dynamic factory fl oor is a potent source of electromagnetic interference (EMI) and vibration. T us: • Sensor data robustness: the sensor streams (especially ToF and Lidar) must be immune to common-mode noise and ambient light saturation. T is requires the implementation of advanced fi ltering and modulation techniques, such as using adaptive Kalman fi lters in the data stream to mitigate vibration eff ects and
Spatial AI and mixed reality fundamentally change the accessibility of industrial automation
Page 1 |
Page 2 |
Page 3 |
Page 4 |
Page 5 |
Page 6 |
Page 7 |
Page 8 |
Page 9 |
Page 10 |
Page 11 |
Page 12 |
Page 13 |
Page 14 |
Page 15 |
Page 16 |
Page 17 |
Page 18 |
Page 19 |
Page 20 |
Page 21 |
Page 22 |
Page 23 |
Page 24 |
Page 25 |
Page 26 |
Page 27 |
Page 28 |
Page 29 |
Page 30 |
Page 31 |
Page 32 |
Page 33 |
Page 34 |
Page 35 |
Page 36 |
Page 37 |
Page 38 |
Page 39 |
Page 40 |
Page 41 |
Page 42 |
Page 43 |
Page 44 |
Page 45 |
Page 46 |
Page 47 |
Page 48
