https://www.mdu.se/

Ensuring mission success for multi-robot systems operating in unpredictable environments requires robust mechanisms to react to unpredictable events, such as robot failures, by adapting plans in real-time. Adaptive mechanisms are especially needed for large teams deployed in areas with unreliable network infrastructure, for which centralized control is impractical and where network segmentation is frequent. This paper advances the state of the art by proposing a field-based runtime task replanning approach grounded in aggregate programming. Through this paradigm, the mission and the environment are represented by continuously evolving fields, enabling robots to make decentralized decisions and collectively adapt the ongoing plan. We compare our approach with a simple late-stage replanning strategy and an oracle centralized continuous replanner. We provide experimental evidence that the proposed approach achieves performance close to the oracle if the communication range is sufficient, while significantly outperforming the baseline even under sparse communication. Additionally, we show that the approach can scale well with the number of robots.