UniLM-Nav: A Unified Framework for Zero-Shot Last-Mile Navigation

*Equal contribution Corresponding author
1Tsinghua University 2BIGAI 3Harbin Institute of Technology 4Peking University

TL;DR: UniLM-Nav is a zero-shot MLLM framework that bridges object navigation and manipulation by selecting task-relevant views, grounding affordances, and reasoning about manipulation-ready base poses for open-vocabulary mobile manipulation.

Loading video...

Watch on YouTube

Abstract

Mobile manipulation requires a robot to navigate to a target object or receptacle and then perform intended manipulation. However, reaching the vicinity of the target does not guarantee a manipulation-ready base pose, a problem known as last-mile navigation. Prior methods for last-mile navigation either rely on manual pose annotation or task-specific training, limiting their scalability to open-vocabulary settings with fine-grained spatial constraints.

We propose UniLM-Nav, a unified framework for zero-shot open-vocabulary last-mile navigation. UniLM-Nav decomposes last-mile navigation into view selection, task-conditioned affordance grounding, and geometry-aware base-pose reasoning, all resolved with a shared multimodal large language model (MLLM) backend. Specifically, UniLM-Nav first selects a reference view that best captures the target object or receptacle from recently collected observations. It then grounds task-relevant affordance point in the selected view and lifts the result into the robot-centric coordinate frame. Finally, conditioned on the grounded affordance, task context, and robot geometry, it infers a manipulation-ready base pose for the robot.

We evaluate UniLM-Nav on the OVMM benchmark where it outperforms the previous state-of-the-art method, MoTo, by 3.13 percentage points. Analyses show that the components of our method are crucial to final performance, and that the choice of MLLM also has a substantial effect. We further deploy UniLM-Nav on a Unitree B2 quadruped robot with a 6-DoF Unitree Z1 manipulator, validating its applicability to real-world mobile manipulation tasks.

UniLM-Nav teaser figure

Method

UniLM-Nav decomposes last-mile navigation into view selection, affordance grounding, and base-pose reasoning. Given a task instruction, the MLLM first selects a suitable view from recent observations, then grounds the task-relevant affordance and lifts it to 3D with depth. Conditioned on the 3D affordance, robot configuration, and task instruction, the MLLM predicts a manipulation-ready base pose for execution.

UniLM-Nav method overview

Results on OVMM-Bench

(1) UniLM-Nav achieves the best performance on the OVMM validation set. (2) With a lightweight RoboBrain-2.5-4B as its MLLM backend, UniLM-Nav still outperforms most alternative methods.

Method Partial Success Rates Overall
SR
Average
SR
FindObj (↑) Pick (↑) FindRec (↑)
HomeRobot (RL) 66.60% 61.10% 50.90% 14.80% 48.30%
HomeRobot (Heuristic) 65.40% 54.80% 43.70% 7.30% 42.80%
MoManipVLA 66.10% 62.60% 53.10% 15.80% 49.40%
UniTeam 66.13% 62.65% 54.69% 17.96% 50.36%
MoTo 66.67% 60.95% 49.87% 20.64% 49.53%
UniLM-Nav (w/ Gemini-3-Flash-Preview) 69.47% 66.22% 54.55% 23.77% 53.50%
UniLM-Nav (w/ RoboBrain-2.5-4B) 68.97% 63.05% 52.38% 19.19% 50.90%

Error Analysis

Left: Error Breakdown. Right: Representative last-mile navigation failures involving view selection, affordance grounding, and base-pose reasoning. The green annotations indicate preferred views, affordance regions, or base poses, while the red annotations show the corresponding erroneous predictions.

Qualitative error analysis of UniLM-Nav on the OVMM benchmark

BibTeX

Copied!
@misc{zhang2026unilmnav,
      title={UniLM-Nav: A Unified Framework for Zero-Shot Last-Mile Navigation},
      author={Zhang, Zhuofan and Wang, Tianxu and Zhang, Guoxi and Lin, Yixiong and Wang, Xilin and Xu, Hongming and Li, Qing and Zhu, Song-Chun and Fan, Lifeng},
      year={2026},
      eprint={2607.06537},
      archivePrefix={arXiv},
      primaryClass={cs.RO},
      url={https://arxiv.org/abs/2607.06537},
}