Multi-robot closed-loop adaptive sampling framework proposed

Shayesteh Hafezi and Alan Papalia propose a multi-robot closed-loop adaptive sampling framework that uses a single learned generative model to simultaneously perform prediction, uncertainty estimation, planning, and measurement assimilation. The model generates ensemble forecasts in latent space and decodes them into physical ocean states, uses ensemble statistics to construct A- or D-optimal sampling objectives, selects points within the robot-reachable range constrained by predicted ocean currents, and feeds measurements back into the latent state through observation-conditioned diffusion sampling. The authors report that across 120 test scenarios, A- and D-optimal sampling generally outperformed static, random, and systematic coverage baselines, with only a small increase in planning overhead; however, the experiments also found that in several scenarios the model uncertainty deviated from forecast error, which the authors list as a direction for future uncertainty quantification work.

AI 根据报道生成 · 6 days ago更新