OpenSLA

OpenSLA provides Sensor-Language-Action models for research using sensor histories and language context. The models produce structured action predictions and sensor-state captions from waveform and numeric inputs. A base variant uses a sensor-token prefix, while a hierarchical variant compresses sensor information into several memory levels. The Python package includes checkpoint loading, a prediction API and a command-line interface that writes JSONL results.

OpenSLA README · Reviewed 2026-11-03