ADAPT: Adaptive Dual-projection Architecture for Perceptive Traversal

2026-03-17

AI Summary

This work addresses the challenge of balancing perceptual accuracy and computational efficiency in humanoid robot locomotion within complex 3D environments, where existing approaches typically rely on fixed perception configurations. The authors propose a dual-projection environmental representation that combines a horizontal elevation map with a vertical distance map, and—critically—introduce the perceptual range as a learnable action for the first time. This enables the policy to adaptively adjust its observation field based on the robot’s motion state, thereby reducing observation dimensionality while enhancing terrain adaptability. Compared to voxel-based baselines, the method substantially lowers computational overhead and accelerates training convergence. It achieves zero-shot transfer to the Unitree G1 humanoid robot and demonstrates exceptional robustness in traversing diverse 3D terrains.

Technology Category

• Intelligent Robots: Multimodal Perception & Sensor Fusion
• Humans and AI: Human-Aware Planning and Behavior Prediction
• Search and Optimization: Learning to Search

Application Category

• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Agile humanoid locomotion in complex 3D en- vironments requires balancing perceptual fidelity with com- putational efficiency, yet existing methods typically rely on rigid sensing configurations. We propose ADAPT (Adaptive dual-projection architecture for perceptive traversal), which represents the environment using a horizontal elevation map for terrain geometry and a vertical distance map for traversable- space constraints. ADAPT further treats its spatial sensing range as a learnable action, enabling the policy to expand its perceptual horizon during fast motion and contract it in cluttered scenes for finer local resolution. Compared with voxel-based baselines, ADAPT drastically reduces observation dimensionality and computational overhead while substantially accelerating training. Experimentally, it achieves successful zero-shot transfer to a Unitree G1 Humanoid and signifi- cantly outperforms fixed-range baselines, yielding highly robust traversal across diverse 3D environtmental challenges.

Problem

• humanoid locomotion
• perceptual fidelity
• computational efficiency
• 3D environments
• adaptive sensing

Innovation

• adaptive perception
• dual-projection representation
• learnable sensing range
• humanoid locomotion
• zero-shot transfer