SuperNav: An Agentic Navigation System for Any Task in Any Scene

Jinkai Zhang; Jingyi Xu; Yuanhong Yu; Jiarui Guo; Ruizhen Hu; Hujun Bao; Xiaowei Zhou; Sida Peng

cs.ROOct 8, 2026

SuperNav: An Agentic Navigation System for Any Task in Any Scene

Authors: Jinkai Zhang, Jingyi Xu, Yuanhong Yu, Jiarui Guo, Ruizhen Hu, Hujun Bao, Xiaowei Zhou, Sida Peng

Organizations: Zhejiang University · Shenzhen University · Causa Robotics

Abstract

General-purpose service robots need navigation systems that can handle diverse human requests in unfamiliar environments, combining task generality with scene generality. Some existing methods fine-tune multimodal large language models (MLLMs) to predict navigation actions, making their behavior dependent on the coverage of navigation training data and potentially limiting generalization to new requests and environments. Our key insight is to let the MLLM focus on interpreting requests, understanding scenes, and making decisions while preserving its general-purpose capabilities and delegating motion execution to navigation tools. To realize this idea, we introduce SuperNav, which equips a pretrained MLLM with a specialized agent harness without navigation-specific fine-tuning of the MLLM. Our harness supports these decisions with Navigation Skills, agent-oriented Tools for physical interaction, and task-progress and context management. A unified visual-point interface connects decision-making to motion by allowing the model to specify destinations directly in images and revise its decisions from execution feedback. Together, these components support sustained navigation across different task requirements and environments. SuperNav outperforms four evaluated baselines on instance-level, multi-object, and demand-driven tasks. Category-level evaluation on HM3D and deployment on a real quadruped robot further demonstrate its applicability across environments. Project Page: https://zju3dv.github.io/SuperNav/

Figures & tables

Figure 1: Skill-guided sofa search. The Agent rechecks candidates, recovers from a blocked doorway, and verifies completion.

Figure 2: Navigation with Skills and agent-oriented Tools. (a) The Agent loop maintains context for MLLM reasoning, Skill loading, and Tool calls. (b) Navigation Skills guide search, candidate rechecking, recovery, and completion; the recovery example shows the Agent retracing its path after a blocked passage. (c) Agent-oriented Tools support observation, motion, goal-progress tracking, and task closure. Point navigation shares an image-space goal interface across geometric and learned execution backends.

	Our instance-navigation benchmark				Demand-driven benchmark
	Single-object (150)		Multi-object (150)		Demand-driven (200)
Method	SR $\uparrow$	SPL $\uparrow$	SR $\uparrow$	SPL $\uparrow$	SR $\uparrow$	SPL $\uparrow$
NaVid	24.67	0.1599	2.67	0.0216	17.00	0.0920
UniNaVid	34.00	0.1833	1.33	0.0131	25.00	0.0943
StreamVLN	13.33	0.1036	0.00	0.0000	25.50	0.0775
OmniNav (Action Former)	27.33	0.2133	4.00	0.0275	37.50	0.1213

Table 1: Single-object, multi-object, and demand-driven navigation results. SR (%) and SPL (0–1): instance tasks use viewpoint geodesic distance

<1

m; demand-driven tasks use 2 m bounding-box regions. Task counts are in parentheses; bold marks column bests.

Figure 3: Ordered multi-object navigation in simulation. The Agent visits a washing machine, a potted plant, and a mahjong table in order.

Figure 4: Demand-driven navigation. To prepare a workspace, the Agent locates a table, identifies a vase to clear, and finds a suitable shelf for it.

	HM3D-OVON val-unseen				HM3Dv2
	0.25 m		1 m		0.2 m		1 m
Method	SR $\uparrow$	SPL $\uparrow$	SR $\uparrow$	SPL $\uparrow$	SR $\uparrow$	SPL $\uparrow$	SR $\uparrow$	SPL $\uparrow$
MTU3D ( Zhu et al., 2025 )	40.80	0.1210	—	—	—	—	—	—
SoftNav ( Wu et al., 2026 )	66.70	0.2570	—	—	—	—	—	—
AstraNav-Memory ( Ren et al., 2025 )	—	—	62.50	0.3490	—	—	—	—
OmniNav (slow + CoT) ( Xue et al., 2026 )	—	—	59.20	0.3320	—	—	—	—

Table 2: Category-level navigation on HM3D. SR (%) and SPL (0–1) at the indicated thresholds. External entries are published results; L3MVN, VLFM, and SG-Nav use ApexNav’s HM3Dv2 reevaluation. Bold marks the highest value in each column.

	0.25 m		1 m
Variant	SR $\uparrow$	SPL $\uparrow$	SR $\uparrow$	SPL $\uparrow$
w/o Navigation Skills	35.00	0.2460	54.17	0.3894
w/o Agent-oriented Tool Design	52.50	0.2230	65.83	0.2855
w/o Skills and Tool Design	31.67	0.1870	54.17	0.3417
Grounding Motion Tool	55.83	0.3201	62.50	0.3624
SuperNav (GPT-6 Astra medium)	71.67	0.4579	79.17	0.4974

Table 3: OVON ablations with Geo-based Executor. The 0.25 m and 1 m thresholds apply to the shortest navigable-path distance from the agent’s final position to the nearest goal viewpoint. Each variant uses 120 episodes and GPT-5.6 Terra (high) unless specified. SR (%) and SPL (0–1); bold marks the best value.

Figure 5: Real-world navigation on Unitree Go2. The robot searches for (a) a basketball and (b) another Go2, and (c) visits a printer followed by a trash bin.

Appendix figures & tables6 assets

Supplementary material from the paper’s appendix.

Appendix

Variant	Entry Skill	Tool-usage Skill
Geo-based Executor	global-navigation-geo-based-executor	localnav-pointnav-geo-based-executor
Learned Executor	global-navigation-learned-executor	localnav-pointnav
Multi-goal	multi-global-navigation-learned-executor	localnav-pointnav

Appendix

Table 9

Multi-object navigation
Method	SR ${}_{1}\uparrow$	SR ${}_{2}\uparrow$	SR ${}_{3}\uparrow$	SR ${}_{4}\uparrow$	SR ${}_{5}\uparrow$
Eligible tasks	150	150	112	75	38
NaVid	40.00	6.67	2.68	1.33	0.00
UniNaVid	36.67	4.00	1.79	1.33	0.00
StreamVLN	22.00	3.33	0.89	0.00	0.00
OmniNav (Action Former)	44.00	11.33	1.79	1.33	0.00

Appendix

Table 4: Ordered progress on multi-object and demand-driven tasks. Multi-object tasks use viewpoint geodesic distance

<1

m; demand-driven tasks use 2 m bounding-box regions. SR k measures completion of the first

k

goals among tasks with at least

k

goals, without requiring STOP. Overall task SR appears in Table 1 .

Method	RGB views	Resolution per view	HFOV	Precision
NaVid	Front	$640\times 480$	$90^{\circ}$	FP16
UniNaVid	Front	$640\times 480$	$120^{\circ}$	FP16
StreamVLN	Front	$640\times 480$	$79^{\circ}$	BF16
OmniNav (Action Former)	Left, front, right	$720\times 640$	$110^{\circ}$	BF16

Appendix

Table 5: Baseline observation settings in Habitat-GS and AI2-THOR. Dimensions refer to environment-rendered images before each model’s preprocessing. Camera height is 1.25 m. All policies receive the complete instruction and retain their native history mechanisms.

Variant	Calls	Navigation	Turns	Path (m)
w/o Navigation Skills	7.43	5.33	0.10	7.56
w/o Agent-oriented Tool Design	30.28	10.72	17.56	23.60
w/o Skills and Tool Design	15.37	5.61	7.72	10.47
SuperNav (Full)	11.17	8.76	0.41	16.44

Appendix

Table 6: Interaction and execution statistics for Geo-based Executor ablations. Means include all 120 episodes, both successes and failures. Calls count Habitat tools, including unsuccessful calls; turns may serve observation or heading adjustment.

Figure 6: Learned Executor architecture. A point selected in a fixed reference image is fused with a bounded, age-aware RGB history. The policy predicts displacement increments and a stop score; the executor converts the increments to motion and returns fresh observations for replanning.

Figure 7: Variable-length visual memory. Each row shows every stride-two candidate on a shared observation-index scale, from the start to the current frame. Colored frames are selected; gray frames are omitted from this fusion pass but remain available in the history. Repeated thinning of the older candidate half limits the input to at most 12 observations, selecting 5, 10, and 11 frames at

t=8

, 24, and 60, respectively.

Explore similar work

ImagiNav: Scalable Embodied Navigation via Generative Visual Prediction and Inverse Dynamics
Mar 14, 2026Jie Chen, Yuxin Cai, Yizhuo Wang +5Robot Navigation Vision-Language Navigation
Department of Mechanical Engineering, National University of Singapore, Singapore · Institute for Infocomm Research (I2R), Agency for Science, Technology and Research (A*STAR), Singapore · Nanyang Technological University (NTU), Singapore
LightNav-0: Eliciting VLM Spatial Intelligence for Generalist Embodied Navigation
Date pendingShaoan Wang, Aocheng Luo, Fei Huang +17Embodied Navigation Cross-Embodiment Robot Learning
Light Origins Team
Qwen-RobotNav Technical Report: A Scalable Navigation Model Designed for an Agentic Navigation System
Jun 16, 2026Jiazhao Zhang, Gengze Zhou, Hale Yin +32Robot Foundation Models Robot Navigation
Qwen Team

Mar 14, 2026, cs.ROJ/K move · Enter open · S save

ImagiNav: Scalable Embodied Navigation via Generative Visual Prediction and Inverse Dynamics

Jie Chen, Yuxin Cai, Yizhuo Wang, Ruofei Bai+4

Department of Mechanical Engineering, National University of Singapore, Singapore · Institute for Infocomm Research (I2R), Agency for Science, Technology and Research (A*STAR), Singapore · Nanyang Technological University (NTU), Singapore

Robot Navigation Vision-Language Navigation Data-Efficient Robot Learning

Open paper AskPDF arXiv

Enabling robots to navigate open-world environments via natural language is critical for general-purpose autonomy. Yet, Vision-Language Navigation has relied on end-to-end policies trained on expensive, embodiment-specific robot data. While recent foundation models trained on vast simulation data show promise, the challenge of scaling and generalizing persists due to the limited scene diversity and visual fidelity in simulation. To address this gap, we propose ImagiNav, a novel hierarchical paradigm that formulates navigation in visual space. Instead of predicting waypoints, ImagiNav synthesizes a future egocentric video conditioned on language instructions, serving as a high-level plan, interpreted by an inverse dynamics model to extract metric trajectories for execution. By decoupling planning from robot actuation, the paradigm enables direct utilization of diverse in-the-wild navigation videos. To support this, we develop an auto-labeling data pipeline that enhances motion annotation accuracy. ImagiNav demonstrates strong zero-shot transfer to robot navigation without requiring robot demonstrations, paving the way for generalist robots that learn navigation directly from unlabeled, open-world data. The project page is available at: https://j1dan.github.io/ImagiNav

Abstract

Explore similar work

ImagiNav: Scalable Embodied Navigation via Generative Visual Prediction and Inverse Dynamics

LightNav-0: Eliciting VLM Spatial Intelligence for Generalist Embodied Navigation

Qwen-RobotNav Technical Report: A Scalable Navigation Model Designed for an Agentic Navigation System