Cambridge researchers' 'Wake Up' model predicts fluid wakes 8x better for robot swarms

A research team from the University of Cambridge and ETH Zurich has published a paper titled 'Wake Up to the Past,' introducing a novel AI model that solves a critical problem for robot swarms: predicting chaotic fluid wake effects. When robots like drones or underwater vehicles move, they create turbulent wakes that can destabilize nearby robots. Traditional data-driven models, which map only the current states of robots to observed forces, fail in agile scenarios because the wake effect has a finite propagation time. The new model's key innovation is incorporating a memory of past robot states and explicitly predicting transport delays, allowing it to accurately forecast the disturbance a 'sufferer' robot will experience.

The researchers empirically tested seven different data-driven model architectures across four different fluid media. They found that models supporting historical state input and transport delay prediction substantially outperformed memory-less alternatives, achieving up to 8x better accuracy. For real-world validation, they built a planar gantry system with two spinning 'monocopters' to collect data and test feedback control. The conclusion is clear: to model the spatio-temporal chaos of fluid interactions, AI must 'wake up to the past.' This work, submitted to IROS 2026, provides a foundational framework for building more robust and tightly coordinated multi-robot systems that operate in fluid environments.