Mobile Robot · Social Navigation

NORM-Nav: Zero-Shot Mobile Robot Navigation with Natural Language Behavioral Constraints

ICRA 2026
Dongjie Huo1*, Junhui Wang2,3*‡, Chao Gao2†, Yan Qiao3, Dong Zhang1, Guyue Zhou2,4†
1 College of Information Science and Technology, Beijing University of Chemical Technology
2 Institute for AI Industry Research (AIR), Tsinghua University
3 Institute of Systems Engineering and Collaborative Laboratory for Intelligent Science and Systems, Macau University of Science and Technology
4 School of Vehicle and Mobility, Tsinghua University
*Equal contribution. †Corresponding authors. ‡Project lead.
Paper (coming soon) arXiv Code

Highlights

Zero-Shot Navigation Framework

A framework that interprets natural language behavioral constraints and incorporates them into costmap-based planning without requiring additional training.

Modular Integration

Seamlessly integrates with standard costmap-based navigation systems without modifying the underlying planner, supporting flexible adaptation to different behavioral requirements.

Comprehensive Evaluation

Validated through extensive simulation and real-world experiments, demonstrating improved task success rates and human-like trajectory generation.

Abstract

Mobile robots operating in human-centered environments must generate not only collision-free paths but also trajectories that follow local behavioral conventions. Conventional costmap-based navigation emphasizes geometric feasibility and often overlooks such requirements, which can result in socially inappropriate behaviors. This paper presents NORM-Nav, a zero-shot framework that integrates natural language behavioral constraints into costmap-based planning. An LLM parses each instruction into structured constraints and grounds them using real-time vision–LiDAR perception. These constraints are encoded as multi-layer costmaps that represent geometric, semantic, directional, and velocity cues and are directly compatible with standard grid-based planners. Simulation and real-world experiments indicate that NORM-Nav improves task success rates and produces trajectories closer to human references than representative baselines.

Behavioral Constraints Navigation

Three navigation examples under language constraints: (a) the robot adjusts its speed, (b) it traverses a perceptually detected curtain, and (c) it bypasses a manhole cover from the left side. Blue lines show executed trajectories, red lines show human reference paths.

Examples of robot navigation under natural language behavioral constraints. The blue line indicates the executed trajectory, while the red line shows a human-preferred reference path. The robot is able to (a) adjust its speed while complying with behavioral constraints, (b) traverse perceptually detected but traversable obstacles such as curtains, and (c) comply with side-specific instructions, such as bypassing a manhole cover from the left even though it is physically traversable.

System Framework

NORM-Nav system architecture diagram: a large language model parses natural language instructions, a vision-LiDAR module grounds them in the scene, and the parsed constraints are encoded into multi-layer costmaps that feed a standard grid-based motion planner.

The architecture of the proposed method for zero-shot navigation under natural language behavioral constraints. The system integrates LLMs with vision-LiDAR perception and encodes parsed behavioral instructions into multi-layer costmaps for motion planning.

Simulation Experiments

Simulation results on three representative navigation tasks, comparing trajectories generated by NORM-Nav against several baselines and a human-operated reference path.

Simulation results on three representative navigation tasks. The proposed method produces stable trajectories that closely follow human-operated reference paths, outperforming baseline approaches.

Real-World Experiments

Real-world demonstrations: the robot follows natural language behavioral instructions without collisions and produces trajectories that closely match human-operated reference paths.

Real-world demonstrations of behavior-constrained navigation. The proposed method successfully follows natural language instructions without collisions, generating stable trajectories that remain close to human-operated reference paths.

Experimental Video

BibTeX

If you find our work useful in your research, please consider citing:

@misc{huo2026normnav,
  title={NORM-Nav: Zero-Shot Mobile Robot Navigation with Natural Language Behavioral Constraints},
  author={Dongjie Huo and Junhui Wang and Chao Gao and Yan Qiao and Dong Zhang and Guyue Zhou},
  year={2026},
  eprint={2605.16979},
  archivePrefix={arXiv},
  primaryClass={cs.RO}
}