HiPAN: Hierarchical Posture-Adaptive Navigation for Quadruped Robots in Unstructured 3D Environments
Quadruped robots can now navigate narrow, height-constrained spaces autonomously.
Researchers from KAIST (Jeil Jeong, Minsung Yoon, et al.) have developed HiPAN (Hierarchical Posture-Adaptive Navigation), a framework that enables quadruped robots to navigate unstructured 3D environments like collapsed buildings, caves, or dense forests. Unlike conventional approaches that rely on sequential mapping and planning pipelines (which suffer from accumulated perception errors and high computational overhead), HiPAN operates directly on onboard depth images at deployment, making it suitable for resource-constrained platforms.
HiPAN adopts a two-tier design: a high-level policy that outputs strategic navigation commands (planar velocity and body posture), and a low-level posture-adaptive locomotion controller that executes those commands. To prevent myopic behaviors and enable long-horizon navigation, the team introduced Path-Guided Curriculum Learning, which progressively extends the navigation horizon from simple obstacle avoidance to strategic goal-reaching. In simulation, HiPAN achieved higher navigation success rates and better path efficiency than classical reactive planners and end-to-end baselines. Real-world experiments across diverse, unstructured 3D environments further validated its effectiveness. The work has been accepted to RA-L 2026.
- HiPAN uses only onboard depth images, eliminating the need for expensive mapping pipelines
- Path-Guided Curriculum Learning extends navigation from reactive avoidance to long-horizon planning
- Outperforms classical reactive planners and end-to-end baselines in both simulation and real-world tests
Why It Matters
Enables practical deployment of quadruped robots in search-and-rescue, inspection, and exploration tasks.