Integrating Proactivity into General-Purpose Agent Frameworks

Develop and evaluate approaches to integrate proactive behavior into general-purpose agent frameworks such as OpenClaw when used with continuous egocentric sensing on smart glasses.

Background

VisionClaw currently operates reactively, acting only when users initiate a task by voice. The authors aim to extend this to proactive behavior that leverages continuous visual and auditory context.

Although prior wearable systems have explored proactivity, the authors explicitly state that it remains unclear how to integrate proactivity with a general-purpose agent backend like OpenClaw, which orchestrates multi-step tool use and personal data access.