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

2026-06-16

AI Summary

This work proposes an extensible navigation model based on the Qwen-RobotNav architecture to address the challenge that existing general-purpose navigation models struggle to flexibly adapt their visual perception strategies to diverse task requirements during inference. The model employs parameterized interfaces to dynamically switch task modes and controllable observation parameters—such as token budget and camera weighting—and leverages randomized configurations during training to enable zero-architecture-change multi-task deployment. This design facilitates composition of complex behaviors by high-level planners, circumventing the reactive degeneration commonly observed in purely trajectory-based training paradigms. Evaluated on standard navigation benchmarks, the model achieves new state-of-the-art performance, demonstrating strong scaling properties, cross-task transferability, and robust zero-shot generalization on real robots.

Abstract

Agentic navigation systems require a base navigation model whose observation strategy can be externally reconfigured at inference time, because instruction following, object search, target tracking, and autonomous driving share the same perception-planning backbone yet demand fundamentally different strategies for consuming the visual stream. We present Qwen-RobotNav, a scalable navigation model built on Qwen-RobotNav that addresses it through a parameterised interface with two complementary dimensions: multiple task modes that select the navigation behaviour, and controllable observation parameters (e.g., token budget, per-camera weights) that govern how visual history is encoded. With training-time randomization over all parameters, Qwen-RobotNav is robust to any inference-time configuration requiring zero architectural modification to the Qwen-RobotNav backbone. We train Qwen-RobotNav on 15.6M samples; co-training with vision-language data prevents the collapse into reactive action-sequence mappers observed in trajectory-only training. The parameterised interface also makes Qwen-RobotNav a natural building block for agentic systems: for long-horizon scenarios, an upper-level planner decomposes goals into sub-tasks and dynamically switches Qwen-RobotNav's task mode and context strategy mid-episode, composing complex behaviours from repeated calls to the same model. Extensive experiments show that Qwen-RobotNav sets new state-of-the-art results across major navigation benchmarks. The model exhibits favourable scaling from 2B to 8B parameters, with joint multi-task training developing a shared spatial-planning substrate that transfers across task families, and demonstrates strong zero-shot generalisation to real-world robots across diverse environments.

Problem

• agentic navigation
• visual observation strategy
• task-specific perception
• navigation model flexibility
• multi-task navigation

Innovation

• parameterized interface
• agentic navigation
• multi-task training
• zero-shot generalization
• scalable navigation model