Energy Vision--Language--Action: A Controlled Multimodal Benchmark for Intent-Conditioned Residential Energy Management
The results show strong asymmetry in modality use: the processed-language pathway is strongly associated with prediction quality, while the current RGB pathway provides no aggregate error advantage, and the results characterize the fixed pilot subset and executed protocol rather than full-benchmark training.
L. S. Saoud, Oualid Doukhi, E. Reihani et al.
· 0 citations