FOCUS: Fine-Grained Open-Vocabulary Change Detection for Uncertainty-Aware Semi-Static Scenes

2026-10-05

AI Summary

This study addresses the challenge of detecting geometrically similar object replacements in semi-static environments by proposing the FOCUS framework. The method fuses 3D Gaussian Splatting-rendered appearance with geometric likelihoods for probabilistic inference, enabling uncertainty-aware, object-level change detection. To robustly handle imperfect rendering evidence, it introduces a recursive three-state estimator and an automatic parameter calibration mechanism based on no-change replay. Experimental results demonstrate that FOCUS significantly improves the replacement detection F1 score from 0.20 to 0.84 on the Isaac Sim benchmark. Furthermore, the framework transfers directly to real-world scenarios without requiring hyperparameter retuning, validating its strong generalization capability.

Abstract

Autonomous robots operating over long periods must keep their environmental memory up to date as the world changes between visits. In semi-static environments, an object may be replaced in place by a different but geometrically and semantically similar instance, making the identity change difficult to detect from geometry or coarse semantics alone. We present FOCUS, an uncertainty-aware framework for object-level change detection and map maintenance. We formulate semi-static memory maintenance as probabilistic inference that fuses geometric likelihood with appearance likelihood rendered from a 3D Gaussian map. A recursive three-state estimator maintains whether each mapped object is PERSISTED, REPLACED, or REMOVED. To account for imperfect 3DGS rendering, we model the rendered appearance evidence probabilistically rather than using it as a direct change score, with the model parameters automatically calibrated from a change-free replay of the initial mapping session. We evaluate our method on a new Isaac Sim warehouse benchmark with ambiguous in-place replacements and on the real-world TorWIC dataset. It improves object-level replacement F1 from 0.20 to 0.84 over a probabilistic baseline and transfers to real-world data without manual parameter retuning.

Problem

• Change Detection
• Semi-Static Environments
• Open-Vocabulary
• Map Maintenance
• Autonomous Robots

Innovation

• Open-Vocabulary Change Detection
• 3D Gaussian Splatting
• Uncertainty-Aware Probabilistic Inference
• Recursive Three-State Estimator
• Semi-Static Scene Memory