Preprint · Under Review

PBD-AG: Persistent Baseline-Delta Active Graphs with Uncertainty-Aware Inspection for Long-Horizon Service Robots

Shuo Bao*Wei Dong*Shuyue ZhangMing Shang Yuchen HuangHan YuChengjie XuYiheng BiKai Sun Fuchun Sun†Xinzhou Wang†

* Equal contribution    † Corresponding authors

▶ Video ◇ Method “ BibTeX
PBD-AG overview: frontier exploration, fixture inspection, grounded RGB-D observations, and baseline-delta scene memory

PBD-AG lets a robot explore once, inspect what matters, and revise object state with evidence—without forgetting the stable structure of the world.

Abstract

Long-horizon service robots require persistent world models that can be built autonomously in unseen environments and revised as task-relevant objects change. Existing methods rely on online mapping, which accumulates localization and observation errors, static scene representations that cannot capture persistent object changes, or holistic vision-language predictions that lack verifiable 3D geometric evidence. We present PBD-AG, a persistent baseline-delta active graph framework that decouples robot-verified stable fixtures from revisable dynamic object events. The robot autonomously bootstraps the structural baseline from onboard exploration and inspects discovered fixtures to ground hierarchical object beliefs. PBD-AG maintains reliability-weighted object states over geometry, semantics, identity, existence, and support relations, using a geometric visibility gate to mitigate false deletions under occlusion. Inspection viewpoints are selected by a graph-conditioned policy that balances target coverage, travel cost, collision risk, and redundant observation. Simulation experiments in multiple environments and under controlled dynamic evaluation show higher aggregate coarse-fixture F1 than capability-matched controls, as well as stronger identity continuity and event recall. A qualitative physical-robot demonstration further illustrates integration with onboard sensing, providing a traceable world model for long-horizon robotic perception.

Core idea. Keep a verified structural baseline, record dynamic changes as event deltas, and only treat a missed detection as negative evidence when the object should have been visible.

Method

A versioned scene graph grounded by active perception.

PBD-AG connects online exploration, fixture-centered inspection, hierarchical grounding, and persistent graph maintenance in one auditable loop.

Modular architecture of PBD-AG
The PBD-AG pipeline transforms embodied RGB-D and LiDAR observations into a versioned stable-fixture baseline and an append-only log of object events.
01

Bootstrap

Explore an unknown environment and discover stable structural fixtures.

02

Inspect

Select reachable and informative fixture-centered viewpoints.

03

Ground

Build hierarchical object beliefs from close-range RGB-D evidence.

04

Revise

Append add, move, reparent, and remove events to persistent memory.

Uncertainty-aware inspection

A miss is not evidence unless the object was observable.

The visibility gate rejects negative evidence when a predicted object lies outside the view frustum, is occluded, or has insufficient expected pixel support. When more evidence is needed, the robot selects only reachable, informative, and safe viewpoints.

Geometric visibility gate and reachable active viewpoint selection

Results

Persistent identity under shared evidence.

PBD-AG improves structural reconstruction and preserves dynamic object identity while reducing false absence updates.

0.868Coarse-fixture F1
0.833Dynamic identity IDF1
11 / 12Event recall
0.014False absence
Coarse fixture F1 comparison across three environments
Highest mean coarse-fixture F1 across all three environments.
Dynamic identity and event recall comparison
Persistent IDs and visibility-gated updates improve identity continuity and event recall.

Demonstrations

From simulation to a physical robot.

The same persistent memory abstraction supports autonomous mapping, inspection, and object-state revision across both settings.

Simulation. Autonomous exploration, inspection, and persistent updates in unseen indoor environments.
Physical deployment. Onboard RGB-D and LiDAR sensing materialize a hierarchical scene graph.

Physical deployment

A traceable world model built from onboard sensing.

Physical robot exploration, inspection, mapping, and hierarchical graph output

BibTeX

@article{bao2026pbd_ag,
  title   = {PBD-AG: Persistent Baseline-Delta Active Graphs with
             Uncertainty-Aware Inspection for Long-Horizon Service Robots},
  author  = {Bao, Shuo and Dong, Wei and Zhang, Shuyue and Shang, Ming and
             Huang, Yuchen and Yu, Han and Xu, Chengjie and Bi, Yiheng and
             Sun, Kai and Sun, Fuchun and Wang, Xinzhou},
  year    = {2026},
  journal = {Preprint, under review}
}