Abstract
Terrain analysis is critical for the practical application of ground mobile robots in real-world tasks, especially in outdoor unstructured environments. In this paper, we propose a novel spatial-temporal traversability assessment method, which aims to enable autonomous robots to effectively navigate through complex terrains. Our approach utilizes sparse Gaussian processes (SGP) to extract geometric features (curvature, gradient, elevation, etc.) directly from point cloud scans. These features are then used to construct a high-resolution local traversability map. Then, we design a spatial-temporal Bayesian Gaussian kernel (BGK) inference method to dynamically evaluate traversability scores, integrating historical and real-time data while considering factors such as slope, flatness, gradient, and uncertainty metrics. GPU acceleration is applied in the feature extraction step, and the system achieves real-time performance. Extensive simulation experiments across diverse terrain scenarios demonstrate that our method outperforms SOTA approaches in both accuracy and computational efficiency. Additionally, we develop an autonomous navigation framework integrated with the traversability map and validate it with a differential driven vehicle in complex outdoor environments.
Introduction
For a mobile robot to navigate an outdoor, unstructured environment—such as a forest or a disaster site—it must determine which areas are "traversable" (safe to drive on) and which are hazardous. Traditional terrain analysis methods often face a "trilemma" of challenges. First, standard Gaussian Processes (GP) provide excellent uncertainty estimates but suffer from high computational complexity ($O(N^3)$), making them too slow for real-time use. Second, simple elevation maps often ignore critical geometric features like local curvature or steep gradients. Finally, many existing sparse methods lack temporal consistency, failing to remember historical observations and leading to flickering maps when sensors are occluded.
System Architecture
To address these challenges, we developed the FSGP-BGK framework. This pipeline is designed to provide high-efficiency terrain assessment by combining feature-based Sparse Gaussian Processes with a Bayesian Gaussian Kernel fusion method.
The system architecture decouples heavy processing tasks. By moving feature extraction (using KNN and PCA) and SGP regression to the GPU, the CPU is freed for high-level path planning and control. This optimization allows the robot to update its local traversability map at a high frequency of 20Hz.
Feature-Based Traversability Assessment
Instead of relying solely on raw elevation data, our method extracts key geometric features directly from raw point clouds. We prioritize curvature and gradient, as these are better proxies for navigability than height alone. By focusing on these "feature points" (e.g., the edges of a cliff) and downsampling redundant flat areas, the SGP can represent complex terrain with a small subset of "inducing points."
The "secret sauce" of the framework is the Spatial-Temporal BGK Fusion. Rather than accumulating massive amounts of old point cloud data, we use a Bayesian Gaussian Kernel to fuse current SGP predictions with a historical memory buffer. This ensures the map remains consistent over time, fills in gaps caused by sensor shadows, and uses a temporal decay weight to prioritize recent observations.
Key Results and Performance
We evaluated FSGP-BGK using a mix of simulated environments (EPFL Terrain Generator, SceneNet) and real-world testing on an Agilex Scout Mini robot.
- Accuracy: In outdoor forest scenarios, FSGP-BGK improved mean accuracy by 18% to 20% over baseline SGP methods. In indoor environments, the improvement reached up to 48%.
- Efficiency: The system achieved an average runtime of 33.84ms, roughly 3x faster than traditional Elevation Map (EM) methods (107.85ms).
- Memory Savings: The BGK fusion method reduced GPU memory consumption from 36% to 25% compared to multi-frame accumulation methods.
Conclusion and Future Work
FSGP-BGK provides a robust, real-time solution for robot navigation in complex terrains. While the current framework focuses on the robot's immediate vicinity, future work will explore integrating this into persistent global SLAM frameworks and extending the approach to multi-robot systems for large-scale exploration. Additionally, explicitly modeling dynamic obstacles remains a key area for future iteration.
Code Availability: The implementation is open-source at https://github.com/ZJU-FAST-Lab/FSGP_BGK.
Summary updated