UCLA OpenSLA Unifies 79 Sensor Modalities, 60 Action Groups and 116K Patients With Language Bridge

UCLA's Sensor-Language-Action framework trains a Qwen-3.5-2B-based OpenSLA with hierarchical sensor encoders across MIMIC-III/IV, operating-room VitalDB and metabolic MetaboNet data covering 116,000+ individuals and 79 sensor modalities. The model generates language-grounded explanations alongside hierarchical action predictions and transfers zero-shot to unseen drugs like Imipenem. The authors explicitly note the model has not undergone clinical validation and is not for direct clinical use.