This repository contains the official implementation of MoTiVE, a generative self-supervised framework for resilient trust management in Vehicular Ad Hoc Networks (VANETs), as presented in the paper:
"MoTiVE: A Mobility and Time-Integrated Graph Autoencoder for Trust Management in VANETs"
Hassan Khaleghirad, Ahmad Khonsari, Mahdi Dolati
Ad Hoc Networks Journal (Elsevier)
MoTiVE addresses the challenges of trust evaluation in highly dynamic VANETs by integrating:
- Mobility-aware positional encoding (MoPE) to capture vehicle trajectory patterns
- ST-EdgeFormer encoder with edge- and time-aware attention
- Multi-level masking (node, edge, and temporal-span) for generative self-supervised learning
- Anomaly-aware trust detection to filter unreliable interactions
Figure: Overall architecture of MoTiVE (from the paper)
- Label-free learning – No ground-truth trust annotations required
- Mobility-aware – Explicitly encodes vehicle positions and velocities via MoPE
- Robust to adaptive attacks – Sybil, bad‑mouthing, good‑mouthing, on‑off, and compound attacks
- Real‑time capable – Sub‑50 ms inference on 3,000 nodes
- State‑of‑the‑art performance – 5.4% AUC improvement over baselines on VeReMi
- Python 3.9 or higher
- CUDA-capable GPU (recommended, but CPU works)
- 16+ GB RAM
pip install -e .