Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

AION: Aerial Indoor Object-Goal Navigation Using Dual-Policy Reinforcement Learning

Zichen Yan, Yuchen Hou, Shenao Wang, Yichao Gao, Rui Huang, Lin Zhao

TL;DR

This work addresses indoor ObjectNav for aerial platforms with 3D locomotion by introducing AION, a dual-policy reinforcement learning framework that separates exploration and goal-reaching. It fuses RGB-D observations, textual target grounding via CLIP, and altitude information through cross-modality attention, and leverages depth-derived cues for obstacle avoidance and ROI-based exploration. The approach yields state-of-the-art results on the AI2-THOR benchmark and demonstrates real-time feasibility and safety in IsaacSim with realistic drone dynamics, highlighting improvements in exploration efficiency, navigation performance, and collision avoidance. The findings suggest that a dual-policy strategy, coupled with depth-aware perception and zero-shot grounding, provides a practical and scalable solution for drone-based indoor ObjectNav in real-world deployments.

Abstract

Object-Goal Navigation (ObjectNav) requires an agent to autonomously explore an unknown environment and navigate toward target objects specified by a semantic label. While prior work has primarily studied zero-shot ObjectNav under 2D locomotion, extending it to aerial platforms with 3D locomotion capability remains underexplored. Aerial robots offer superior maneuverability and search efficiency, but they also introduce new challenges in spatial perception, dynamic control, and safety assurance. In this paper, we propose AION for vision-based aerial ObjectNav without relying on external localization or global maps. AION is an end-to-end dual-policy reinforcement learning (RL) framework that decouples exploration and goal-reaching behaviors into two specialized policies. We evaluate AION on the AI2-THOR benchmark and further assess its real-time performance in IsaacSim using high-fidelity drone models. Experimental results show that AION achieves superior performance across comprehensive evaluation metrics in exploration, navigation efficiency, and safety. The video can be found at https://youtu.be/TgsUm6bb7zg.

AION: Aerial Indoor Object-Goal Navigation Using Dual-Policy Reinforcement Learning

TL;DR

This work addresses indoor ObjectNav for aerial platforms with 3D locomotion by introducing AION, a dual-policy reinforcement learning framework that separates exploration and goal-reaching. It fuses RGB-D observations, textual target grounding via CLIP, and altitude information through cross-modality attention, and leverages depth-derived cues for obstacle avoidance and ROI-based exploration. The approach yields state-of-the-art results on the AI2-THOR benchmark and demonstrates real-time feasibility and safety in IsaacSim with realistic drone dynamics, highlighting improvements in exploration efficiency, navigation performance, and collision avoidance. The findings suggest that a dual-policy strategy, coupled with depth-aware perception and zero-shot grounding, provides a practical and scalable solution for drone-based indoor ObjectNav in real-world deployments.

Abstract

Object-Goal Navigation (ObjectNav) requires an agent to autonomously explore an unknown environment and navigate toward target objects specified by a semantic label. While prior work has primarily studied zero-shot ObjectNav under 2D locomotion, extending it to aerial platforms with 3D locomotion capability remains underexplored. Aerial robots offer superior maneuverability and search efficiency, but they also introduce new challenges in spatial perception, dynamic control, and safety assurance. In this paper, we propose AION for vision-based aerial ObjectNav without relying on external localization or global maps. AION is an end-to-end dual-policy reinforcement learning (RL) framework that decouples exploration and goal-reaching behaviors into two specialized policies. We evaluate AION on the AI2-THOR benchmark and further assess its real-time performance in IsaacSim using high-fidelity drone models. Experimental results show that AION achieves superior performance across comprehensive evaluation metrics in exploration, navigation efficiency, and safety. The video can be found at https://youtu.be/TgsUm6bb7zg.
Paper Structure (28 sections, 17 equations, 6 figures, 5 tables)

This paper contains 28 sections, 17 equations, 6 figures, 5 tables.

Table of Contents

  1. INTRODUCTION
  2. Related Works
  3. Zero-Shot ObjectNav
  4. Aerial ObjectNav on Drones
  5. Proposed Methods
  6. Task Definition
  7. Cross-Modality Attention
  8. Spatial Perception From Depth
  9. Depth Projection for Obstacle Avoidance
  10. Depth ROI for Exploration
  11. Goal-Reaching Policy Training
  12. Approaching Reward
  13. Parent Reward
  14. Bounding Box Reward
  15. Exploration Policy Training
  16. ...and 13 more sections

Figures (6)

  • Figure 1: Demonstration of two-stage aerial ObjectNav with policy switching between exploration (AION-e) and goal-reaching (AION-g).
  • Figure 2: Overview of the proposed dual-policy RL framework for aerial indoor ObjectNav.
  • Figure 3: Process of spatial perception from depth images.
  • Figure 4: ObjectNav trajectories of AION-g across different iTHOR rooms.
  • Figure 5: Placement of unseen objects at varying heights across different indoor environments in IsaacSim.
  • ...and 1 more figures