Collision-Aware Traversability Analysis for Autonomous Vehicles in the Context of Agricultural Robotics

TL;DR

This work tackles safe autonomous navigation for agricultural UGVs in unstructured environments by fusing LiDAR depth with multispectral imagery to build a mass-density augmented map. It introduces a physics-based traversability metric where the expected velocity after potential collisions is captured by $v_R^f = \frac{m_R}{m_R+m_i} v_R$ and the continuous cost $\alpha = \exp\left(-\frac{1}{m_R}\int_{\mathcal{P}} d_m(a)\, da\right)$ (discretized as $\alpha = \exp\left(-\frac{1}{m_R} \sum_{c_i\in\mathcal{P}} d_m(i)\, a_c\right)$). Mass density per semantic class is computed as $\mathbb{E}[d_m] = \frac{\sum_i d_m(c_i)\, p(s|c_i)}{\sum_i p(s|c_i)}$, enabling a 2D traversal-cost grid used by a local planner to select safer paths. The approach is validated with a real-world sensor setup, showing NDVI-based vegetation indices offer robust segmentation, and demonstrates how semantic awareness reduces risky interactions while remaining computationally feasible for real-time deployment. Overall, the method provides a collision-aware, physics-grounded framework for adaptable, semantically informed traversability in agricultural robotics.

Abstract

In this paper, we introduce a novel method for safe navigation in agricultural robotics. As global environmental challenges intensify, robotics offers a powerful solution to reduce chemical usage while meeting the increasing demands for food production. However, significant challenges remain in ensuring the autonomy and resilience of robots operating in unstructured agricultural environments. Obstacles such as crops and tall grass, which are deformable, must be identified as safely traversable, compared to rigid obstacles. To address this, we propose a new traversability analysis method based on a 3D spectral map reconstructed using a LIDAR and a multispectral camera. This approach enables the robot to distinguish between safe and unsafe collisions with deformable obstacles. We perform a comprehensive evaluation of multispectral metrics for vegetation detection and incorporate these metrics into an augmented environmental map. Utilizing this map, we compute a physics-based traversability metric that accounts for the robot's weight and size, ensuring safe navigation over deformable obstacles.

Figures (5)

• Figure 1: Autonomous plateform navigating through an agricultural field. Tall grass, as well as crops, protrude into the path, forming potential obstacles the robot has to collide with. As such, any system working in such conditions must be able to assess the severity of each collision.
• Figure 2: Traversability analysis flow: Multispectral images are fused with LiDAR data to produce an augmented point cloud, which serves as the basis for semantic segmentation. Mass density is calculated for each depth measurement, updating the 3D environmental map. This is then converted into a 2D traversability grid. Ultimately, potential paths are assessed, and the safest one is chosen.
• Figure 3: Augmented 3D maps of a park environment, consisting of tall grass, bushes, trees and a small shack. Top Left: Color camera's image from the scene; Top Right: ndvi colorized 3D map; Bottom Left: visible light colorized 3D map; Bottom Right: Manually annotated 3D map
• Figure 4: Reflectance profiles of several elements of agricultural environment
• Figure 5: Local planning candidates evaluated based on the mass density grid. Semantic grid (top) and satellite view (middle) illustrate the environmental configuration during the run (middle). The local paths over the cost grid (bottom) illustrate navigation costs of each candidate. A coefficient of $1$ means no loss of velocity, thus a safe path, whereas a coefficient of $0$ means a collision resulting in the total stop of the robot.