TY - JOUR
T1 - EDOG-Map
T2 - Evidential Dynamic Occupancy Grid Mapping for Consumer-Grade UAV Navigation in Dynamic Environments
AU - Tan, Zheng
AU - Zhang, Zuowei
AU - Lyu, Yang
AU - Pan, Quan
N1 - Publisher Copyright:
© 1975-2011 IEEE.
PY - 2026
Y1 - 2026
N2 - While grid mapping is foundational for the safe navigation of unmanned aerial vehicles (UAVs), traditional binary Bayes filters inherently assume a static world, rendering them inadequate for highly dynamic environments. Conversely, advanced particle-driven Random Finite Set (RFS) frameworks capture dynamic states but impose prohibitive computational and memory overheads, precluding their deployment on Size, Weight, and Power (SWaP)-constrained platforms. To bridge this gap, we propose EDOG-Map, an efficient Evidential Dynamic Occupancy Grid mapping framework. EDOG-Map explicitly models spatial uncertainty utilizing Dempster-Shafer (D-S) theory. By introducing a hierarchical evidential architecture, our method extracts dynamic states directly from the instantaneous logical conflict masses between temporal belief layers. This bypasses the need for complex velocity distribution sampling or prior object models, enabling instantaneous, cell-level moving obstacle detection. Furthermore, a robot-centric local map coupled with a constant-time memory update strategy guarantees scalability in large-scale scenes, preserving high-resolution perception while drastically minimizing the memory footprint. Extensive evaluations on public benchmarks and real-world UAV flights demonstrate that EDOG-Map achieves robust, minimal-ghosting dynamic detection with an F1-Score exceeding 90%. Crucially, it reduces computational latency by over 50% and peak memory consumption by up to 82% compared to state-of-the-art baselines, ensuring the highly consistent processing latency required for safe, high-speed autonomous flight.
AB - While grid mapping is foundational for the safe navigation of unmanned aerial vehicles (UAVs), traditional binary Bayes filters inherently assume a static world, rendering them inadequate for highly dynamic environments. Conversely, advanced particle-driven Random Finite Set (RFS) frameworks capture dynamic states but impose prohibitive computational and memory overheads, precluding their deployment on Size, Weight, and Power (SWaP)-constrained platforms. To bridge this gap, we propose EDOG-Map, an efficient Evidential Dynamic Occupancy Grid mapping framework. EDOG-Map explicitly models spatial uncertainty utilizing Dempster-Shafer (D-S) theory. By introducing a hierarchical evidential architecture, our method extracts dynamic states directly from the instantaneous logical conflict masses between temporal belief layers. This bypasses the need for complex velocity distribution sampling or prior object models, enabling instantaneous, cell-level moving obstacle detection. Furthermore, a robot-centric local map coupled with a constant-time memory update strategy guarantees scalability in large-scale scenes, preserving high-resolution perception while drastically minimizing the memory footprint. Extensive evaluations on public benchmarks and real-world UAV flights demonstrate that EDOG-Map achieves robust, minimal-ghosting dynamic detection with an F1-Score exceeding 90%. Crucially, it reduces computational latency by over 50% and peak memory consumption by up to 82% compared to state-of-the-art baselines, ensuring the highly consistent processing latency required for safe, high-speed autonomous flight.
KW - Consumer-grade UAVs
KW - Dynamic environments
KW - Environmental perception
KW - Evidential theory
KW - Occupancy grid mapping
UR - https://www.scopus.com/pages/publications/105048528429
U2 - 10.1109/TCE.2026.3725136
DO - 10.1109/TCE.2026.3725136
M3 - 文章
AN - SCOPUS:105048528429
SN - 0098-3063
JO - IEEE Transactions on Consumer Electronics
JF - IEEE Transactions on Consumer Electronics
ER -