GoalVLM: VLM-driven Object Goal Navigation for Multi-Agent System
Fuente:
arXiv
Saved in:
| Main Authors: | , , , |
|---|---|
| Format: | Preprint |
| Published: |
2026
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866911528303722496 |
|---|---|
| author | James, MoniJesu Habel, Amir Atef Fedoseev, Aleksey Tsetserokou, Dzmitry |
| author_facet | James, MoniJesu Habel, Amir Atef Fedoseev, Aleksey Tsetserokou, Dzmitry |
| contents | Object-goal navigation has traditionally been limited to ground robots with closed-set object vocabularies. Existing multi-agent approaches depend on precomputed probabilistic graphs tied to fixed category sets, precluding generalization to novel goals at test time.
We present GoalVLM, a cooperative multi-agent framework for zero-shot, open-vocabulary object navigation. GoalVLM integrates a Vision-Language Model (VLM) directly into the decision loop, SAM3 for text-prompted detection and segmentation, and SpaceOM for spatial reasoning, enabling agents to interpret free-form language goals and score frontiers via zero-shot semantic priors without retraining. Each agent builds a BEV semantic map from depth-projected voxel splatting, while a Goal Projector back-projects detections through calibrated depth into the map for reliable goal localization. A constraint-guided reasoning layer evaluates frontiers through a structured prompt chain (scene captioning, room-type classification, perception gating, multi-frontier ranking), injecting commonsense priors into exploration.
We evaluate GoalVLM on GOAT-Bench val_unseen (360 multi-subtask episodes, 1032 sequential object-goal subtasks, HM3D scenes), where each episode requires navigating to a chain of 5-7 open-vocabulary targets. GoalVLM with N=2 agents achieves 55.8% subtask SR and 18.3% SPL, competitive with state-of-the-art methods while requiring no task-specific training. Ablation studies confirm the contributions of VLM-guided frontier reasoning and depth-projected goal localization. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2603_18210 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | GoalVLM: VLM-driven Object Goal Navigation for Multi-Agent System James, MoniJesu Habel, Amir Atef Fedoseev, Aleksey Tsetserokou, Dzmitry Robotics Object-goal navigation has traditionally been limited to ground robots with closed-set object vocabularies. Existing multi-agent approaches depend on precomputed probabilistic graphs tied to fixed category sets, precluding generalization to novel goals at test time. We present GoalVLM, a cooperative multi-agent framework for zero-shot, open-vocabulary object navigation. GoalVLM integrates a Vision-Language Model (VLM) directly into the decision loop, SAM3 for text-prompted detection and segmentation, and SpaceOM for spatial reasoning, enabling agents to interpret free-form language goals and score frontiers via zero-shot semantic priors without retraining. Each agent builds a BEV semantic map from depth-projected voxel splatting, while a Goal Projector back-projects detections through calibrated depth into the map for reliable goal localization. A constraint-guided reasoning layer evaluates frontiers through a structured prompt chain (scene captioning, room-type classification, perception gating, multi-frontier ranking), injecting commonsense priors into exploration. We evaluate GoalVLM on GOAT-Bench val_unseen (360 multi-subtask episodes, 1032 sequential object-goal subtasks, HM3D scenes), where each episode requires navigating to a chain of 5-7 open-vocabulary targets. GoalVLM with N=2 agents achieves 55.8% subtask SR and 18.3% SPL, competitive with state-of-the-art methods while requiring no task-specific training. Ablation studies confirm the contributions of VLM-guided frontier reasoning and depth-projected goal localization. |
| title | GoalVLM: VLM-driven Object Goal Navigation for Multi-Agent System |
| topic | Robotics |
| url | https://arxiv.org/abs/2603.18210 |