GROVE: A Training-Free Framework for Proactive Video Memory and Reasoning
A new framework called GROVE has been developed by researchers to improve wearable assistants. This innovative system allows these devices to answer inquiries regarding their visual history while also discerning when that information is pertinent to present circumstances. In contrast to traditional video-memory systems that focus on question-based recall, GROVE offers proactive support through a singular memory that evolves from a continuous video feed. It captures detailed perceptual data, gradually organizing it into time-stamped instances, coherent narratives, and recurring patterns across days. Both reactive question-answering and proactive assistance utilize the same memory interface, differing only in initiation. GROVE has been tested against various benchmarks, proving its capability to function effectively without additional training. The research paper can be found on arXiv with the identifier 2608.02392.
Key facts
- GROVE is a training-free framework for video memory and reasoning.
- It supports both reactive question-answering and proactive assistance.
- Memory is grown causally from a continuous video stream.
- Memory is organized into time-stamped moments, episodes, and cross-day patterns.
- Each memory stratum has a scale-native retrieval skill.
- Reactive and proactive behaviors share the same memory and access interface.
- The framework was evaluated across multiple benchmarks.
- The paper is available on arXiv (2608.02392).
Entities
Institutions
- arXiv