OVMAN: A Task and Benchmark for Open-Vocabulary Motion-Aware Navigation

Dibyendu Ghosh

cs.ROSep 6, 2026

OVMAN: A Task and Benchmark for Open-Vocabulary Motion-Aware Navigation

Authors: Dibyendu Ghosh

Abstract

Homes change between a robot's visits. Navigation benchmarks pose their goals in the world the agent currently sees, and the two-visit benchmarks that exist score recall or rearrangement rather than navigation. None of them can express go to the chair that was moved or go to where the vase used to be. OVMAN is a task in which an agent tours a scene, returns after a scripted change, and must navigate to a goal specified by the change itself. Two of its six change relations answer with a place an object has left, where nothing remains to be detected. We release 219 two-visit episodes, each certified solvable by an oracle agent that completes it three times. Two released systems fail as predicted. A zero-shot object-goal navigator reaches the change-defined target in 12.3% of episodes and almost never reaches a vacated location, 0.026 on former and 0.050 on removed. When a self-maintaining open-vocabulary map is read as two visits rather than as one maintained map, overall success rises (0.396 to 0.479 on our mapping) but the past-position relations specifically do not recover, on our maps or on a released self-maintaining one; keeping a map current is therefore not the whole obstacle. A simple two-visit reference agent reaches 45.2% when navigating, against an embodied oracle of 99.5%. An error decomposition places the remaining difficulty in carrying an instance identity across visits rather than in naming it or in selecting the answer once positions are known.

Figures & tables

	two	open	goal by	empty past
	visits	vocab.	change rel.	locations
RoomR [ 1 ]	yes	no	no	no
EMQA [ 2 ]	yes	no	no	no (no nav.)
FindingDory [ 3 ]	yes	yes	no	no
3D-SMNet [ 4 ]	yes	no	partly	no (no nav.)
SuperMap [ 23 ]	yes	yes	no	no (history only)

TABLE I: OVMAN against the closest prior work. No individual column is new; no other row occupies all four.

Fig. 1: The six relations on recorded episodes from the released set, one column each: the agent’s view on the familiarisation tour (top) and from the same tour after the change (bottom), with the sentence it receives. The pin marks the answer, a floor-plane position: a filled star where the answer is an object that must be visible at the stop, a hollow star where the answer is a place the object has left. Frames are selected automatically by projecting the answer into the recorded views, none by hand.

Fig. 2: The reference agent. Both per-visit maps are retained (dashed box); the two highlighted stages, cross-visit association and the change relations derived from it, are what a single-state navigation stack or a current-state maintained map does not provide directly.

symbol	meaning	value
$W$	waypoints per tour (4 headings each)	6
$R$	success radius	1.0 m
$\Delta_{\min}$	minimum applied displacement	1.5 m
$\gamma$	association gate	1.25 m
$\lambda$	feature-disagreement scale	2.0 m
$\tau$	moved / stable displacement threshold	0.75 m

TABLE II: All constants of the benchmark and the reference agent, as read from the released code.

grounding handed over (diagnostic)
system	memory	MOVED	FORMER	MOVING	STABLE	APPEARED	REMOVED	ALL
ours (reference)	single visit	0.230	0.149	0.145	0.223	0.237	0.142	0.186
DualMap [ 17 ]	maintained map	0.202	0.158	0.139	0.198	0.133	0.183	0.167
DualMap	two visits	0.222	0.193	0.056	0.323	0.292	0.175	0.209
ours (reference)	two visits	0.545	0.447	0.361	0.281	0.375	0.475	0.416
DualMap	maintained map	0.274	0.275	0.321	0.325	0.513	0.395	0.355

TABLE III: Static gate: the predicted point must fall within

1.0

m of the answer. “Grounding handed over” replaces a system’s own language grounding with ground truth over its own map objects. DualMap is not run-to-run deterministic, so its rows are means over three runs; ours are deterministic.

agent	MVD	FMR	MVG	STB	APP	RMV	SR	SPL
CoW [ 27 ]	0.121	0.026	0.194	0.219	0.150	0.050	0.123	—
ours, 1 visit	0.424	0.132	0.389	0.375	0.325	0.100	0.283	0.269
ours, 2 visits	0.576	0.421	0.472	0.406	0.425	0.425	0.452	0.440
ours, oracle pos	1.000	0.974	0.972	1.000	1.000	1.000	0.991	0.982
oracle check	1.000	1.000	0.972	1.000	1.000	1.000	0.995	0.990

TABLE IV: Embodied evaluation: the agent navigates with the primitives of Table II . CoW runs from its released code [ 27 ] .

outcome	MVD	FMR	MVG	STB	APP	RMV	ALL
success	18	17	13	9	15	19	91
abstain (no candidate)	9	10	12	2	17	17	67
elsewhere ( $>2$ m)	4	10	5	17	5	3	44
near miss (1–2 m)	1	0	3	3	2	0	9
wrong instance	1	1	3	1	1	1	8
episodes	33	38	36	32	40	40	219

TABLE V: How the two-visit agent fails (static gate). “Wrong instance”: within

R

of another instance of the category. The first five rows partition the

219

episodes; “lost in navigation” is a subset of success — static successes lost when executed.

Fig. 3: Static success of the two-visit agent (a) by the number of instances of the queried category (123, 37 and 59 episodes), with 95% intervals, against the single-visit baseline and the same agent with grounding handed over; (b) by scene.

Fig. 4: Two visits against one current-state reading, per relation, with grounding handed over: our mapping (left), the released DualMap (right). The gain is on the relations that still leave an object to find; on former and removed it is negligible or reversed.

Fig. 5: (a) Error budget on the static gate; each rung hands over one error source. (b) Static success against the association gate

\gamma

: the span over

[0.75,2.5]

m is

0.027

, and the value used lies below

\Delta_{\min}

by construction.

Robot Navigation

Robot navigation studies how robots perceive, localize, plan, and move safely toward goals in static or changing environments.

See all 532 papers

SAP-Nav: Spatial Semantic Representation Meets Active Perception for Hierarchical Open-Vocabulary Object Navigation
Aug 13, 2026Xuetong Pei, Jian Liu, Vidura Munasinghe +4Active Perception Embodied Navigation
1Beihang University · 2Singapore University of Technology and Design · 3AIML, Adelaide University
OmniNav: Robust Long-Horizon Target Navigation in Dynamic Environments
Sep 8, 2026Yujie Tang, Meiling Wang, Jinhao Jiang +4Robot Navigation Object-Goal Navigation
School of Automation, Beijing Institute of Technology, Beijing, China.
DRIVE-Nav: Directional Reasoning, Inspection, and Verification for Efficient Open-Vocabulary Navigation
Mar 30, 2026Maoguo Gao, Zejun Zhu, Zhiming Sun +6Embodied Navigation Open-Vocabulary Object Navigation
1Beijing Institute of Technology · 2DeepBlue College · 3The National Key Laboratory of Autonomous Intelligent Unmanned Systems (KAIUS), School of Automation

Aug 13, 2026J/K move · Enter open · S save

SAP-Nav: Spatial Semantic Representation Meets Active Perception for Hierarchical Open-Vocabulary Object Navigation

Xuetong Pei, Jian Liu, Vidura Munasinghe, Bo Miao+3

1Beihang University · 2Singapore University of Technology and Design · 3AIML, Adelaide University

Active Perception Embodied Navigation Open-Vocabulary Object Navigation Vision-Language Navigation Semantic Mapping

Open paper AskPDF arXiv

Hierarchical open-vocabulary object navigation (OVON) requires agents to follow free-form instructions that may specify targets through scene-, room-, region-, and instance-level cues in unseen environments. Although recent work LangMap has formalized this setting, reliably solving it under partial observations remains challenging: spatial grounding requires persistent environment-level evidence, whereas target verification requires clear and discriminative candidate views. We present SAP-Nav, a fully online, zero-shot framework that addresses both requirements through active perception. SAP-Nav incrementally constructs a Queryable Spatial-Semantic Representation from actively acquired room views, enabling spatial semantic queries from any explored location. It further employs Active Viewpoint Verification to assess whether the current observation provides sufficient evidence and, when necessary, reposition the agent to a more informative viewpoint before verifying candidates against category and attribute constraints. Although designed for hierarchical OVON, SAP-Nav supports both hierarchical and standard category-level OVON without task-specific training or precomputed scene maps. Experiments on LangMap and HM3D-OVON show that SAP-Nav achieves the overall best performance, including a 12.2% improvement in SR over training-based methods on region-level navigation. Real-world robot experiments further demonstrate its practical feasibility. Code will be made publicly available upon acceptance.

Abstract

Robot Navigation

Explore similar work

SAP-Nav: Spatial Semantic Representation Meets Active Perception for Hierarchical Open-Vocabulary Object Navigation

OmniNav: Robust Long-Horizon Target Navigation in Dynamic Environments

DRIVE-Nav: Directional Reasoning, Inspection, and Verification for Efficient Open-Vocabulary Navigation