TAS-TsC: A Data-Driven Framework for Estimating Time of Arrival Using Temporal-Attribute-Spatial Tri-space Coordination of Truck Trajectories

TL;DR

This work tackles truck ETA prediction under challenging GPS data conditions, notably temporal sparsity, variable-length sequences, and inter-vehicle interactions. It introduces TAS-TsC, a tri-space framework that fuses temporal embeddings from a selective State Space Model (Mamba), structured attribute embeddings from statistical feature engineering, and spatial embeddings via a Graph Diffusion-based fusion of trajectories, with a self-supervised training objective and a downstream Histogram-based Gradient Boosting predictor. The approach demonstrates superior accuracy and robustness on Shenzhen data, including strong domain-generalization when transferring across urban regions, and gains are attributed to the complementary tri-space representations and graph-based diffusion of spatial information. Practically, TAS-TsC offers scalable ETA improvements with potential for real-time deployment and multi-modal logistics expansion, supported by rigorous ablations and parameter analyses that illuminate the contribution of each module and guidance for tuning in diverse settings.

Abstract

Accurately estimating time of arrival (ETA) for trucks is crucial for optimizing transportation efficiency in logistics. GPS trajectory data offers valuable information for ETA, but challenges arise due to temporal sparsity, variable sequence lengths, and the interdependencies among multiple trucks. To address these issues, we propose the Temporal-Attribute-Spatial Tri-space Coordination (TAS-TsC) framework, which leverages three feature spaces-temporal, attribute, and spatial-to enhance ETA. Our framework consists of a Temporal Learning Module (TLM) using state space models to capture temporal dependencies, an Attribute Extraction Module (AEM) that transforms sequential features into structured attribute embeddings, and a Spatial Fusion Module (SFM) that models the interactions among multiple trajectories using graph representation learning.These modules collaboratively learn trajectory embeddings, which are then used by a Downstream Prediction Module (DPM) to estimate arrival times. We validate TAS-TsC on real truck trajectory datasets collected from Shenzhen, China, demonstrating its superior performance compared to existing methods.

Figures (8)

• Figure 1: Illustration of coordination of truck GPS trajectories in complex logistics scenarios.
• Figure 2: The overall structure of TAS-TsC framework. TLM captures temporal patterns in trajectory data using state-space modeling. AEM creates structured embeddings from trajectory attributes like speed and direction through feature engineering. SFM builds a spatiotemporal graph to model interactions among trajectories. DPM combines the outputs from TLM, AEM, and SFM using embedded and structure learning, and then uses HGB to predict truck arrival times.These modules work together to enhance ETA accuracy by leveraging temporal, attribute, and spatial data.
• Figure 3: Heat-map of attribute embedding correlation.
• Figure 4: Illustration of GPS track recording locations of trucks in Shenzhen.
• Figure 5: GPS track and statistics of trucks in Shenzhen.