Demo

Overview

Object-Goal Navigation (ObjectNav) requires an agent to autonomously explore unknown environments and navigate toward target objects specified by semantic labels. While prior work has primarily studied zero-shot ObjectNav under 2D locomotion, extending it to aerial platforms with 3D locomotion remains underexplored.

AION is an end-to-end dual-policy reinforcement learning framework that decouples exploration and goal-reaching behaviors into two specialized policies. It enables vision-based aerial ObjectNav without relying on external localization or global maps.

Key Features

  • Dual-Policy RL: Separate specialized policies for exploration and goal-reaching
  • Vision-Based: RGB-D perception without external localization or global maps
  • 3D Locomotion: Designed for aerial robots with superior maneuverability
  • Sim-to-Real: Evaluated on AI2-THOR and IsaacSim with high-fidelity drone models

Resources