Hiking in the Wild: A Scalable Perceptive Parkour Framework for Humanoids

TL;DR

The paper tackles robust humanoid hiking in unstructured environments, addressing the limitations of reactive proprioception and drift-prone exteroception. It proposes Hiking in the Wild, an end-to-end policy that ingests high-frequency depth and proprioception, uses a Mixture-of-Experts backbone, depth-noise simulation, Terrain Edge Detection with Foot Volume Points, and Flat Patch Sampling to prevent reward hacking, enabling zero-shot Sim-to-Real transfer. Key contributions include a scalable perception-planning-control loop, a robust safety mechanism for footholds on edges, and a patch-based navigation target strategy validated by real-world demonstrations reaching speeds up to 2.5 m/s on stairs, slopes, and gaps. The framework is open-source, enabling reproducible research and deployment with minimal hardware changes, and demonstrates practical impact for agile, autonomous humanoid locomotion in the wild.

Abstract

Achieving robust humanoid hiking in complex, unstructured environments requires transitioning from reactive proprioception to proactive perception. However, integrating exteroception remains a significant challenge: mapping-based methods suffer from state estimation drift; for instance, LiDAR-based methods do not handle torso jitter well. Existing end-to-end approaches often struggle with scalability and training complexity; specifically, some previous works using virtual obstacles are implemented case-by-case. In this work, we present \textit{Hiking in the Wild}, a scalable, end-to-end parkour perceptive framework designed for robust humanoid hiking. To ensure safety and training stability, we introduce two key mechanisms: a foothold safety mechanism combining scalable \textit{Terrain Edge Detection} with \textit{Foot Volume Points} to prevent catastrophic slippage on edges, and a \textit{Flat Patch Sampling} strategy that mitigates reward hacking by generating feasible navigation targets. Our approach utilizes a single-stage reinforcement learning scheme, mapping raw depth inputs and proprioception directly to joint actions, without relying on external state estimation. Extensive field experiments on a full-size humanoid demonstrate that our policy enables robust traversal of complex terrains at speeds up to 2.5 m/s. The training and deployment code is open-sourced to facilitate reproducible research and deployment on real robots with minimal hardware modifications.

Figures (10)

• Figure 1: System overview. Our framework trains an end-to-end policy using simulated depth and proprioception. To ensure safety and agility on complex terrains, we incorporate Scalable Edge Penalization to avoid risky footholds and Position-based Command generation for precise tracking. The trained policy is directly deployed to the real robot (Zero-shot) using only a 60 Hz onboard depth camera as exteroception, achieving extremely high-dynamic locomotion without explicit localization or map reconstruction.
• Figure 2: Depth processing: Top row shows $\mathcal{F}_{sim}$ on synthetic data, and bottom row shows $\mathcal{F}_{real}$ on real-world data.
• Figure 3: Automatically detected edges across diverse terrains.
• Figure 4: Volume points distributed within the foot manifold.
• Figure 5: Flat patches on different terrains.