É«ÖÐÉ«

Traditionally, robots operate by analyzing shapes and distances—what’s around them and how close it is. But this geometric view misses the ‘why’ behind what happens. In the real world, semantics—what things mean—help humans navigate safely and efficiently. A pedestrian expects cars to stay off the sidewalk because society has embedded that rule in our traffic system. Similarly, a robot that understands this semantic context can behave more predictably and appropriately. Instead of constantly reacting to every pedestrian, a robot car could focus only on the people who are relevant to its current task.

From traffic to tactics: three applications

This idea is explored by through three progressive studies. The first takes inspiration from traffic systems, linking semantic rules (like traffic signs) directly to a robot’s motion behavior. By encoding these rules into a semantic map, the robot learns how to move in ways that reduce the chance of surprise. In the second study, the robot is faced with unclear route instructions—like when someone forgets to mention a turn. Here, it uses knowledge about spatial relationships between places to adjust its route, similar to how humans correct mistakes on the fly. Finally, in a soccer-playing scenario, robots are programmed with strategic knowledge from team sports. Rather than blindly reacting to player positions, they move with intent, recognizing when an opponent’s formation offers a tactical opportunity or when it’s time to fall back and keep possession.

Designing smarter, more flexible robots

The research shows that semantics aren’t just helpful—they’re essential for real-world robot behavior. They provide structure and expectations, just like they do for people navigating a busy city. Designing adaptable robot behavior becomes possible when motion and task constraints are built as flexible modules, which can be adjusted as situations change. Importantly, building these systems requires collaboration. A soccer coach might contribute knowledge about game tactics, while a robot engineer focuses on how the machine moves. By separating and modularizing this knowledge, experts from different fields can each shape a piece of the robot’s behavior.

Toward truly intelligent machines

In the end, this work proves that semantic knowledge makes robots smarter. It led to a robot that drives more safely using traffic-like rules, one that can recover from vague directions, and soccer robots that don’t just follow fixed patterns, but understand when and how to change their play. It's a step toward robots that not only function in the human world—but truly belong in it.

The research took place in the robotics section of the Mechanical Engineering section at É«ÖÐÉ«. The research falls under the project. More information on the EAISI FAST Lab.

Title of PhD thesis: . Supervisors: Prof. Herman Bruyninckx, Prof. René van de Molengraft, and Dr. Elena Torta.