Robot swarms trained in computer simulations successfully transfer skills to the real world
Researchers at ULB have solved the frustrating 'reality gap' in robotics by limiting what AI can control.

Original image temporarily unavailable
Computer simulations are the safest place to teach robots new tricks, but skills learned virtually usually fall apart when machines face real-world friction, sensor noise, and mechanical imperfections, a persistent hurdle known as the reality gap. According to a study published in Science Robotics by Ilyes Gharbi, Mauro Birattari, and their team at IRIDIA, the artificial intelligence laboratory of ULB, this failure happens because overly flexible reinforcement learning algorithms overfit the simulator, learning quirks that do not exist in reality.
To fix this, the researchers restricted the learning algorithm. Rather than letting it control each wheel's speed directly, they forced it to choose moment by moment among a handful of pre-tested behaviors, such as moving toward a light, aggregating with peers, or exploring. This leaves the algorithm to focus only on what matters: deciding when to switch between these behaviors so the group coordinates effectively. In tests involving a swarm of 20 e-puck robots across 250 physical experiments and five collective missions, swarms transferred to reality reliably and significantly outperformed state-of-the-art methods that tried to control motors directly.
View selection rationale
Importance 75/100
About this brief
- Content type
- Research summary
- Source type
- Publisher report
- Original source
- Robohub ↗
- Source published
- Process
- AI-assisted discovery, translation, summarisation, classification and scoring; standards and corrections are maintained by the editorial desk.
- Verification
- The original URL and publication metadata are retained. This brief is not independently fact-checked unless explicitly stated.
- Published by
- Polyemergence Editorial Desk · PolyEmergence Pte Ltd.
- Conflict disclosure
- No financial or commercial conflict declared.