Which Objects Stay? Learning Hierarchical Persistence Beliefs for Lifelong Open-Vocabulary Object Maps
Which Objects Stay? Learning Hierarchical Persistence Beliefs for Lifelong Open-Vocabulary Object Maps
This research presents HiPer, a hierarchical persistence filter that learns object survival probabilities directly from data without relying on semantic categories. By retrieving visually and geometrically similar training instances, HiPer initializes a conjugate Beta belief over an object's survival probability and updates it using detections and missed observations. When an object is re-identified at a new location, its persistence belief transfers with the object, enabling the system to learn object-specific mobility over time. Experiments on 120 held-out simulated sequences show that HiPer consistently outperforms competing methods, with retrieval-based priors surpassing CLIP-derived label priors. Results on 47 held-out 3RScan environments further demonstrate that the proposed approach exceeds state-of-the-art baselines in real-world settings.
The paper introduces HiPer, a hierarchical persistence filter that uses retrieval-based priors to estimate object survival in open-vocabulary settings, achieving superior performance on both simulated and real-world benchmarks compared to class-based and CLIP-derived baselines.
Source: arXiv Robotics — research abstracts · Read original article ↗
Loading article text…
What the source reports
Publisher-reported claims, with original evidence. These results have not been independently verified by RoboSignal.
Reported numbers
simulated sequences
120
Reported trials
View original evidence
When an object is re-identified at a new location, its persistence belief is transferred with the object, enabling the map to learn object-specific mobility over time. Experiments on 120 held-out simulated sequences show that HiPer consistently achieves the best performance among competing methods. Retrieval-based priors outperform CLIP-derived label priors, and object-specific online learning provides further gains.
Open source S4environments
47
Reported trials
View original evidence
Results on 47 held-out 3RScan environments demonstrate that the proposed retrieval-based prior also outperforms state-of-the-art baselines in real-world settings.
Open source S5
- paper: not_reported
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
When an object is re-identified at a new location, its persistence belief is transferred with the object, enabling the map to learn object-specific mobility over time. Experiments on 120 held-out simulated sequences show that HiPer consistently achieves the best performance among competing methods. Retrieval-based priors outperform CLIP-derived label priors, and object-specific online learning provides further gains.
Open source S4
Results on 47 held-out 3RScan environments demonstrate that the proposed retrieval-based prior also outperforms state-of-the-art baselines in real-world settings.
Open source S5
Which Objects Stay? Learning Hierarchical Persistence Beliefs for Lifelong Open-Vocabulary Object Maps
Open source S1
arXiv:2611.07662v1 Announce Type: cross Abstract: Lifelong object maps must estimate whether an object remains at its previously observed location after it leaves view. Existing persistence filters rely on class-based survival priors, making them difficult to apply reliably in open-vocabulary settings. We introduce HiPer, a hierarchical persistence filter that derives persistence beliefs directly from data rather tha
Open source S2
n semantic categories. For a new object, HiPer retrieves visually and geometrically similar training instances and estimates an initial survival rate from their observed histories, without requiring class labels. This estimate initializes a conjugate Beta belief over the object's survival probability, which is updated in closed form using detections and missed observations under Rosen's persistence-filter framework.
Open source S3
Source：arXiv Robotics — research abstracts · arxiv.org