Neural networks halve distance measurement errors
19 December 2025 
The German sensor-maker Leuze claims that it has been able to cut measurement errors in demanding industrial applications by using neural networks. The technique improves measurement accuracy without needing any extra computing resources.
Optical distance sensors using time-of-flight (TOF) technology offer many benefits. They allow fast, contactless measurement of long distances, are insensitive to ambient light, and provide continuous distance data in real time. They work by recording the time it takes for emitted light – usually in the form of laser or LED pulses – to travel to an object and back.
But TOF technology also has its limitations. Its accuracy depends heavily on the object surface. Dark surfaces can weaken the reflected signal, resulting in narrower pulses and delayed echo detection. Bright surfaces, on the other hand, generate stronger signals with a wider pulse width, that are detected earlier. That means that the returning signal is detected at different times depending on whether the object surface is light or dark, resulting in measurement errors.
Until now, mathematical models based on defined algorithms have been used to correct these errors. A correction value is calculated for many different surfaces and distances, and is later applied automatically. This calculation is based on a polynomial function which is effective for stable, continuous error curves. However, it can result in limited imaging accuracy if there are complicating factors, such as strongly varying surface reflections. Because the model parameters are fixed, the functions cannot adapt automatically to changing environmental conditions.
Leuze says it now has a much more precise and flexible approach. Instead of working with rigid formulae, it is using neural networks to determine the correction value. A neural network is a form of AI that is modelled on the human brain. It consists of nodes (neurons) in three types of layers: the input layer, hidden layers and the output layer. The network processes information by passing input data through these one layer at a time. An “activation function” decides how strongly a neuron becomes active – what value it passes on to the next layer. This function allows the network to learn even complex, non-linear relationships, and is not limited to simple calculations.
Leuze’s AI system uses sample data to learn how brightness and surface texture affect a distance sensor’s measurements. This makes it much easier to correct the measured values. The neural network is trained with data consisting of raw distance values and pulse widths as input parameters, as well as standardised correction values at the output. The training data can be generated from the production process, in which measured values are collected for light, dark and differently textured surfaces, as well as for different distances. These values are communicated to the control system. A neural network calculates the correction values for the sensor from this data. The sensor requires no extra computing power during operation – the AI has already “learned” everything.
Tests have shown that the AI-based calibration technique can reduce the dependence of measurement results on surface and distance by more than half. It results in more robust and accurate measurements without any extra effort during operation, even with difficult surfaces. The benefits include:
- fewer measurement errors, resulting in much more precise results;
- the flexibility to operate with different types of sensor and surfaces;
- improved learning from real data, even with strongly oscillating 3D curve characteristics;
- no need for additional computing during operation; and
- future-proof operation, thanks to its use of AI
Leuze expects typical applications to include:
- navigation and collision avoidance on robots and mobile platforms
- materials handling checking positions and distances on conveyor belts
- quality assurance measuring distances to workpieces with difficult surfaces
- AGVs precise distance control when parking and manoeuvring
- safety detecting proximity to machines and systems
