GSAT: Geometric Traversability Estimation using Self-supervised Learning with Anomaly Detection for Diverse Terrains

Abstract

Safe autonomous navigation requires reliable estimation of environmental traversability. Traditional methods have relied on semantic or geometry-based approaches with human-defined thresholds, but these methods often yield unreliable predictions due to the inherent subjectivity of human supervision. While self-supervised approaches enable robots to learn from their own experience, they still face a fundamental challenge: the positive-only learning problem. To address these limitations, recent studies have employed Positive-Unlabeled (PU) learning, where the core challenge is identifying positive samples without explicit negative supervision. In this work, we propose GSAT, which addresses these limitations by constructing a positive hypersphere in latent space to classify traversable regions through anomaly detection without requiring additional prototypes (e.g., unlabeled or negative). Furthermore, our approach employs joint learning of anomaly classification and traversability prediction to more efficiently utilize robot experience. We comprehensively evaluate the proposed framework through ablation studies, validation on heterogeneous real-world robotic platforms, and autonomous navigation demonstrations in simulation environments.

Figures (8)

• Figure 1: Overall process of the proposed GSAT framework: (a) Automated data generation using positive and unlabeled samples to enable self-supervised learning, corresponding to the initial latent space (i); (b) Self-supervised anomaly detection and hypersphere refinement to optimize the decision boundary within the latent space (ii); (c) Final traversability mapping for autonomous navigation.
• Figure 2: Overview of the proposed GSAT framework. The framework consists of (A) automated data generation leveraging robot traversal supervision, and (B) a traversability network that extracts BEV features to produce latent representations. The core innovation lies in (C) experience-aware traversability learning, where self-supervised anomaly detection in the latent space partitions unlabeled data into normal and anomalous samples, enabling joint traversability estimation and positive hypersphere refinement through the proposed loss formulations.
• Figure 3: Point augmentation methods demonstrated on RELLIS-3D dataset. (a) Sample point clouds. (b) Flipping augmentation reflecting points across the $yz$-plane. (c) Yaw rotation augmentation around the $z$-axis. (d) Pitch augmentation process: (d.1) RANSAC-based ground segmentation, (d.2) slope angle extraction, (d.3) pitch transformation to simulate terrain inclinations.
• Figure 4: Qualitative results of the data augmentation ablation study on the RELLIS-3D dataset, demonstrating spatial prediction performance across different augmentation configurations.
• Figure 5: Experimental setup for traversability mapping evaluation. (a) Robot platforms: SCOUT MINI (left) and Go2 (right). (b) Data collection trajectories in the outdoor environment with training and mapping evaluation paths.