Uncertainty-aware Semantic Mapping in Off-road Environments with Dempster-Shafer Theory of Evidence

TL;DR

The paper tackles unreliable semantic predictions in off-road environments by integrating Evidential Deep Learning with Dempster-Shafer Theory to produce uncertainty-aware semantic maps. It introduces an evidential segmentation stage to quantify semantic belief and uncertainty, and an evidential mapping stage that fuses evidences using reduced Dempster's rule while expanding influence through an adaptive spatial kernel. The approach yields more reliable uncertainty maps and competitive semantic accuracy across off-road datasets, addressing perceptual challenges and enabling safer downstream decisions. The combination of EDL and DST with an extended belief mechanism provides a principled framework for uncertainty propagation and information fusion in mapping under uncertainty.

Abstract

Semantic mapping with Bayesian Kernel Inference (BKI) has shown promise in providing a richer understanding of environments by effectively leveraging local spatial information. However, existing methods face challenges in constructing accurate semantic maps or reliable uncertainty maps in perceptually challenging environments due to unreliable semantic predictions. To address this issue, we propose an evidential semantic mapping framework, which integrates the evidential reasoning of Dempster-Shafer Theory of Evidence (DST) into the entire mapping pipeline by adopting Evidential Deep Learning (EDL) and Dempster's rule of combination. Additionally, the extended belief is devised to incorporate local spatial information based on their uncertainty during the mapping process. Comprehensive experiments across various off-road datasets demonstrate that our framework enhances the reliability of uncertainty maps, consistently outperforming existing methods in scenes with high perceptual uncertainties while showing semantic accuracy comparable to the best-performing semantic mapping techniques.

Figures (3)

• Figure 1: Overview of our evidential semantic mapping framework. With an evidential segmentation network trained by EDL 10_EDL_sensoy2018evidential, input data is processed to derive continuous semantic belief and uncertainty. These 3D evidential points are then integrated into the semantic map through adaptive evidential fusion using Dempster's rule of combination, resulting in a dependable semantic map and uncertainty map in environments with high perceptual uncertainties.
• Figure 2: Qualitative results of 3D semantic mapping methods. Compared to others, our uncertainty-aware mapping methods (EBSkim2024evidential and Ours) generate accurate maps that preserve semantic details while excluding noisy predictions in both datasets. This is depicted by the plane of grass in the first row and the boundaries of unpaved roads, grass, and vegetation in the subsequent rows.
• Figure 3: Visualization of uncertainty maps from 3D semantic mapping methods. The color of each cell represents the relative uncertainty value throughout each map. Though EBS and ours only provide qualitatively meaningful uncertainty maps, EBS tends to overestimate uncertainty. In contrast, our method effectively distinguishes between areas with high and low uncertainty.