Paper Overview
Field: Computer Vision Authors: Shravan Chaudhari, William Paul, Suchi Saria, Rama Chellappa, Homanga Bharadhwaj Published: 2026-10-07 arXiv: 2610.10538
Summary
As we move through the world and carry out everyday tasks, we encounter objects that may become relevant only later — we can recall where we left something or what was inside a container, even without knowing we would need that information. This work studies how an embodied assistant can build similar memory from egocentric videos by observing a person's day-to-day activities.
Ledger: Persistent 3D Object Memory
Ledger combines object locations, their histories, and contextual descriptions. Key properties:
- Associates observations across the whole recording and retains objects after they leave view — including objects the person never touches
- Clusters each object's observations by resting locations, and records a move only after repeated evidence, reducing the effect of localization noise
- Short descriptions preserve details such as object contents or supporting surfaces
- Stored records allow answering spatial questions later without access to original images or video
Results
| Benchmark | Before | After | |---|---|---| | HD-EPIC accuracy | 29.7% | 42.6% | | UCS-Bench accuracy | 33.8% | 38.5% | | Ego4D object localization | — | median error 0.99 m |
--- *Collected automatically on 2026-10-09*