Skip to main content
Dogeartech

Algorized and ZHAW Study Sensor Fusion for Human-Aware Collaborative Robots

A two-year project will combine radar, Wi-Fi and camera data to estimate people’s position and movement near robots, with an under-100 ms inference target.

Algorized and ZHAW Study Sensor Fusion for Human-Aware Collaborative Robots

Collaborative robots need more than a signal that a person is nearby. To adjust their movement, a controller needs an estimate of where that person is going and how reliable the estimate is. A two-year research program from Algorized and the ZHAW Zurich University of Applied Sciences Institute of Signal Processing and Wireless Communications (ISC) will investigate how to provide that information by combining wireless and optical sensor data.

Algorized X ZHAW
Algorized X ZHAW

Called “Real-Time Human Perception via Edge-AI Wireless Sensor Fusion for Collaborative Robotics,” the program began in September 2026 and is scheduled to run for two years. InnoSuisse is funding the work. It builds on earlier research by Algorized and ISC involving ultra-wideband (UWB) radar.

Combining signals before making a decision

The researchers plan to combine raw wireless and optical signals in a fusion layer between the robot’s sensors and its controller. Rather than relying only on separate sensor outputs, the system is intended to weigh inputs according to confidence and produce a calibrated estimate of a person’s location and likely path.

That design puts attention on where sensor fusion happens. Algorized currently processes raw data with one foundation model to detect presence and classify whether a detected object is human. The project will examine whether combining raw inputs before individual sensors make their own decisions can give human-aware machines a more unified view of the surrounding space.

The sensor types contribute different kinds of information. UWB radar and Wi-Fi sensing can help identify a person behind an obstruction such as a shelf or machine. Millimeter-wave radar can maintain detection through conditions including dust, steam, glare and darkness. RGB and time-of-flight cameras contribute spatial and semantic detail. The research will investigate how those signals can complement one another, including when an individual modality has limited visibility or confidence.

Confidence matters because a robot’s response depends on the quality of its perception. A position estimate and a predicted path are useful only if the controller can gauge how much to rely on them. The planned fusion layer is therefore intended to pass both a human movement estimate and information about its reliability toward the control system.

From detection to a graded robot response

The program has four stated goals: detect people approaching a robot, distinguish living from non-living objects, track each person and predict their path, then send the controller a graded instruction to continue, slow down or stop. Algorized’s planned safety functions cover presence detection, localization and tracking.

The target for inference is under 100 milliseconds on standard industrial edge hardware. That timing target makes local processing a central part of the work: the system is meant to deliver its estimate on equipment used near the robot, within the response window the project is aiming to achieve. The target is a research objective, rather than a reported result.

Marc Kuhn, deputy head of ISC at the ZHAW School of Engineering, said the program will bring ISC sensing research into real industrial sites. He described the work as combining radar, Wi-Fi and optical signals to distinguish approaching people from other objects, estimate their position and movement, measure estimation reliability and predict where they may go next.

Training data and transfer between sites

Building a model for this task requires training data collected in real facilities, where people move through working environments. The researchers also need to interpret those measurements with an understanding of the underlying sensing physics. That combination matters because the same observed signal can carry different information depending on the sensor and the conditions around it.

The stated aim is a core model that can transfer across sensor types, vendors and sites without retraining for each deployment. If the project can support that goal, integrators could face less site-specific model work when adapting a perception system to a new robot cell or sensor arrangement. The program has not yet reported transfer performance; establishing whether that approach works is part of the research.

Algorized CEO and founder Natalya Lopareva said the collaboration is intended to explore sensors and compute layers that help robots understand people sharing their workspace. For engineers, the practical question is whether the fused estimate can remain timely and dependable as people and sensing conditions change. The two-year study is set up to examine that question in industrial settings, with a defined response target and a focus on confidence as well as position and motion.

Source: Hannah Shapiro · www.massrobotics.org

Filed alongside