TANGO: Traversablility-Aware Navigation with Local Metric Control for Topological Goals

Stefan Podgorski*1, Sourav Garg*1, Mehdi Hosseinzadeh1, Lachlan Mares1,
Feras Dayoub1, Ian Reid1,2
1The University of Adelaide, 2MBZUAI,
* denotes equal contribution
Paper Code

Presented at ICRA 2025, Atlanta, USA.

TANGO

Abstract

Abstract—Visual navigation in robotics traditionally relies on globally-consistent 3D maps or learned controllers, which can be computationally expensive and difficult to generalize across diverse environments. In this work, we present a novel RGB- only, object-level topometric navigation pipeline that enables zero-shot, long-horizon robot navigation without requiring 3D maps or pre-trained controllers. Our approach integrates global topological path planning with local metric trajectory control, allowing the robot to navigate towards object-level sub- goals while avoiding obstacles. We address key limitations of previous methods by continuously predicting local trajectory using monocular depth and traversability estimation, and in- corporating an auto-switching mechanism that falls back to a baseline controller when necessary. The system operates using foundational models, ensuring open-set applicability without the need for domain-specific fine-tuning. We demonstrate the effectiveness of our method in both simulated environments and real-world tests, highlighting its robustness and deployability. Our approach outperforms existing state-of-the-art methods, offering a more adaptable and effective solution for visual navigation in open-set environments

Method Pipeline

TANGO’s Navigation Pipeline. Perception: The robot’s current view is segmented using a foundational segmentation model (SAM), the segments are localised within an object-level topological map using local feature matching (LightGlue). Each segment is assigned a cost based on its topological proximity to the final goal segment, the segment closest to the final goal is selected to drive the controller. Control: A BEV traversability map is computed by combining state-of-the-art depth estimation with open-set text query capabilities (CLIP) to identify traversable surfaces such as ‘floor’ or ‘ground’. This depth and semantic information is integrated to generate a BEV cost map. Dijkstra’s algorithm is applied to compute the shortest path to the sub- goal segment, providing a trajectory that avoids obstacles and generates yaw control signals for robot navigation. This perception-action loop is repeated continuously until the robot reaches the final goal object.

TANGO

We present a topometric navigation pipeline that uniquely bridges topological global path planner and metric local trajectory planning, without needing 3D maps or learnt controllers. This enables our method to effective avoid obstacles (bottom row) even when no such objects were present in the mapping (teach) run.

Varying Obstacles

TANGO

Successful control traversing around sofa via sub-goals to an originally unseen goal.

TANGO

Failure case - stuck behind a chair with subgoal visible behind

TANGO

BibTeX


        @inproceedings{podgorski2025tango,
          title={TANGO: Traversablility-Aware Navigation with Local Metric Control for Topological Goals},
          author={Podgorski, Stefan and Garg, Sourav and Hosseinzadeh, Mehdi and Mares, Lachlan and Dayoub, Feras and Reid, Ian},
          booktitle={2025 IEEE International Conference on Robotics and Automation (ICRA)},
          year={2025},
          organization={IEEE}
        }