Accepted · IEEE RA-L 2026

HiPAN: Hierarchical Posture-Adaptive Navigation for Quadruped Robots in Unstructured 3D Environments

Jeil Jeong1,*  ·   Minsung Yoon1,*  ·   Seokryun Choi1  ·   Heechan Shin1  ·   Taegeun Yang1  ·   Sung-eui Yoon1,†
1School of Computing, KAIST
*Equal contribution  ·  Corresponding author

Real-world experiment in an outdoor environment.

HiPAN teaser: quadruped navigates dead-end corridor and crouches under overhanging passage.

HiPAN in action. Using only onboard depth, the robot backtracks after entering a dead-end corridor (A, B) and adaptively adjusts its posture to pass under a height-constrained passage (C). The voxel map is shown for visualization only.

Abstract

Navigating quadruped robots in unstructured 3D environments requires goal-directed motion, exploration to escape local minima, and posture adaptation to traverse narrow, height-constrained spaces. Conventional 3D mapping–planning pipelines suffer from accumulated perception errors and high computational overhead on resource-constrained platforms.

We propose HiPAN, a hierarchical framework that operates directly on onboard depth images at deployment. A high-level policy issues strategic navigation commands—planar velocity and body posture—executed by a low-level, posture-adaptive locomotion controller. To mitigate myopic behavior, we introduce Path-Guided Curriculum Learning (PGCL), progressively extending the navigation horizon so the robot first masters reactive avoidance and later develops long-horizon strategies. HiPAN outperforms classical and end-to-end RL baselines in simulation and transfers robustly to diverse indoor and outdoor environments.

Key Contributions

01 · FRAMEWORK

Hierarchical Posture-Adaptive Control

High-level policy outputs velocity and body posture commands directly from depth; low-level policy executes them with adaptive whole-body motion.

02 · LEARNING

Path-Guided Curriculum Learning

Progressively removes intermediate sub-goals along a privileged path, extending the navigation horizon and fostering non-greedy exploration.

03 · SIM RESULTS

96.5% SR · 88.7 SPL

Averaged across four unstructured 3D benchmarks, outperforming classical (56.1% / 41.7) and end-to-end (53.7% / 48.4) baselines.

04 · REAL WORLD

Deployment on Unitree Go1

Onboard RealSense D435i and NUC; robust across cluttered indoors, dead-end corridors, abrupt illumination changes, and outdoor sunlight.

Framework Overview

HiPAN framework overview diagram.

Two-level hierarchy trained stage-wise. Teacher policies (red) are optimized with privileged inputs via PPO; student policies (cyan) are distilled with DAgger and rely only on onboard depth and proprioception at deployment. The high-level policy emits a 5-D command c = [vx, vy, ωz, h, θx], executed by the posture-adaptive low-level controller.

Path-Guided Curriculum Learning

Path-guided curriculum learning diagram.

Early stages provide dense sub-goals along a privileged global path (A), teaching reactive avoidance. As curriculum level increases, sub-goals thin out (B, C) until only the final goal remains (D), forcing the agent to learn long-horizon, non-greedy behavior.

Simulation Results

Success Rate (SR) and Success weighted by Path Length (SPL) across four benchmarks; mean over 10 seeds.

Method Corridor Room Complex-1 Complex-2
SR%SPL SR%SPL SR%SPL SR%SPL
HiPAN (Ours) 98.593.2 98.488.8 94.489.3 94.783.6
HiPAN w/o Posture 98.493.1 94.574.1 74.269.1 73.166.2
HiPAN w/o PGCL 97.888.7 76.969.0 78.168.0 74.566.5
Bug Algorithm 91.158.9 88.167.4 23.820.1 21.420.4
Wall-Following 51.931.7 84.559.1 22.419.8 20.219.7
Flat-RL 82.071.5 43.839.7 45.342.1 43.740.1
Qualitative trajectory comparison across four benchmarks.

Qualitative trajectories. HiPAN (red) reliably reaches the goal, whereas Bug (green), Wall-Following (blue), and Flat-RL (magenta) frequently detour or become trapped.

Simulation video. Rollouts across the four procedurally generated benchmarks — Corridor, Room, Complex-1, and Complex-2 — spanning dead-ends, enclosed rooms, overhanging obstacles, and cluttered layouts.

Real-World Validation

Real-world experiments across four scenarios.

Robot trajectories (cyan) from start (circle) to goal (star) overlaid on LiDAR maps (visualization only). Four scenarios validate posture adaptation, dead-end recovery, illumination robustness, and outdoor generalization.

Scenario 1 — Indoor Clutter

Posture-adaptive navigation among tables, ladders, and box piles.

Scenario 2 — Dead-End Recovery

Backtracking from a dead-end corridor, then continuing toward the goal.

Scenario 3 — Illumination Changes

Robust operation under abrupt lighting loss at a doorway.

Scenario 4 — Outdoors

Direct sunlight, uneven terrain, and unseen obstacles (bicycles, cones).

BibTeX

@misc{jeong2026hipanhierarchicalpostureadaptivenavigation,
      title={HiPAN: Hierarchical Posture-Adaptive Navigation for Quadruped Robots in Unstructured 3D Environments},
      author={Jeil Jeong and Minsung Yoon and Seokryun Choi and Heechan Shin and Taegeun Yang and Sung-eui Yoon},
      year={2026},
      eprint={2604.26504},
      archivePrefix={arXiv},
      primaryClass={cs.RO},
      url={https://arxiv.org/abs/2604.26504}
    }