Digital Twin Technology Helps Robots Forecast and Prevent Workplace Accidents

In modern factories, robots are no longer locked away behind cages. They work side by side with people, lifting parts, assembling products and repeating precise tasks all day long. While this collaboration boosts productivity, it also raises an important question: how can we keep workers safe when machines move so quickly?

A research team from the University of Verona in Italy believes the answer lies not just in reacting to danger, but in predicting it. Their new system can estimate when and where a collision between a human and a robot might happen, before it actually occurs.

Teaching Robots to "Think Ahead"

Most current safety systems focus on avoidance. If a person gets too close, the robot slows down or stops. These systems rely on sensors that measure distance or detect sudden contact. They work, but often at the last possible moment.

The Verona researchers took a different approach. Instead of only measuring how close a robot is to a person right now, their system studies what the robot is likely to do next. Industrial robots usually follow repetitive routines, such as pick-and-place movements. Even if these routines are not directly shared with safety systems, they can be observed and learned over time.

By watching the robot's movements and tracking its internal operation modes, the system gradually builds a model of its behavior. This allows it to forecast future movements and calculate how long it might take before a robot arm reaches a worker's hand or shoulder.

In simple terms, the robot's routine becomes predictable, and that predictability improves safety.

A Digital Twin of the Work Cell

At the heart of the technology is something called a "digital twin." This is a real-time virtual model of both the human worker and the robot.

To track the worker, the team used multiple depth cameras mounted high above the workspace. These cameras capture color and depth information and convert it into a 3D skeleton model of the person. The system identifies key body points, like shoulders, elbows, and hands, and updates their positions several times per second.

At the same time, the robot's joints and internal task variables are monitored. The system records whether the robot is in a working phase or paused, and what part of its task cycle it is performing.

All this information feeds into the digital twin, which constantly mirrors the real environment. The twin can then simulate future robot movements and check whether any part of the robot's predicted path will intersect with the worker's body.

Predicting Collisions in Real Time

To make collision checks fast and reliable, the researchers represent arms and robot links as simple capsule-shaped volumes. This makes it easier for the system to calculate distances between body parts and detect possible overlaps.

If the robot's predicted future path intersects with a worker's position, the system calculates how much time remains before contact. The result is not just a warning that something is close, but an estimate of "time to collision."

A color-based heat map makes the prediction easy to understand. Body parts are shown in cool colors when safe, and gradually shift toward red as the risk increases. This visual feedback can help supervisors or automated safety controls respond early.

Importantly, the system is designed to run in real time. Instead of recalculating everything from scratch whenever a worker moves slightly, it reuses previous calculations whenever possible. This keeps processing fast enough for busy industrial environments.

Safer and Smarter Factories

The research team tested their system using a KUKA industrial robot in a series of static and dynamic scenarios. In some tests, a worker's hand remained in the robot's path. In others, the worker moved around freely. The system successfully adapted to both situations, predicting collisions when necessary and showing no risk when the worker stepped away.

The key advantage of this approach is foresight. Rather than simply stopping a robot when someone gets too close, factories could slow movements, adjust paths, or reorganize tasks before a dangerous situation develops.

As robots become more common in shared workspaces, predictive safety tools like this could play a crucial role. By learning robot routines and combining them with real-time human tracking, the system moves beyond reaction and toward anticipation.

In the future, the researchers hope to extend the technology to handle multiple workers and even forecast human motion. If successful, factories of tomorrow may not just be automated, they may also be far better at protecting the people who work inside them.