Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

DutyTTE: Deciphering Uncertainty in Origin-Destination Travel Time Estimation

Xiaowei Mao, Yan Lin, Shengnan Guo, Yubin Chen, Xingyu Xian, Haomin Wen, Qisen Xu, Youfang Lin, Huaiyu Wan

TL;DR

DutyTTE tackles origin-destination travel time uncertainty by separating the problem into accurate path prediction and segment-level uncertainty modeling. It introduces a deep reinforcement learning framework to maximize overall alignment between predicted and ground-truth paths, and a mixture-of-experts module that captures context-dependent travel-time variability across road segments, calibrated with MIS losses. The approach yields improved travel-time estimates and tighter, more reliable prediction intervals, validated on two real-world taxi datasets with clear gains over strong baselines. By providing calibrated confidence bounds and efficient inference, DutyTTE offers practical benefits for ride-hailing, logistics, and traffic-aware planning under uncertain travel times.

Abstract

Uncertainty quantification in travel time estimation (TTE) aims to estimate the confidence interval for travel time, given the origin (O), destination (D), and departure time (T). Accurately quantifying this uncertainty requires generating the most likely path and assessing travel time uncertainty along the path. This involves two main challenges: 1) Predicting a path that aligns with the ground truth, and 2) modeling the impact of travel time in each segment on overall uncertainty under varying conditions. We propose DutyTTE to address these challenges. For the first challenge, we introduce a deep reinforcement learning method to improve alignment between the predicted path and the ground truth, providing more accurate travel time information from road segments to improve TTE. For the second challenge, we propose a mixture of experts guided uncertainty quantification mechanism to better capture travel time uncertainty for each segment under varying contexts. Additionally, we calibrate our results using Hoeffding's upper-confidence bound to provide statistical guarantees for the estimated confidence intervals. Extensive experiments on two real-world datasets demonstrate the superiority of our proposed method.

DutyTTE: Deciphering Uncertainty in Origin-Destination Travel Time Estimation

TL;DR

DutyTTE tackles origin-destination travel time uncertainty by separating the problem into accurate path prediction and segment-level uncertainty modeling. It introduces a deep reinforcement learning framework to maximize overall alignment between predicted and ground-truth paths, and a mixture-of-experts module that captures context-dependent travel-time variability across road segments, calibrated with MIS losses. The approach yields improved travel-time estimates and tighter, more reliable prediction intervals, validated on two real-world taxi datasets with clear gains over strong baselines. By providing calibrated confidence bounds and efficient inference, DutyTTE offers practical benefits for ride-hailing, logistics, and traffic-aware planning under uncertain travel times.

Abstract

Uncertainty quantification in travel time estimation (TTE) aims to estimate the confidence interval for travel time, given the origin (O), destination (D), and departure time (T). Accurately quantifying this uncertainty requires generating the most likely path and assessing travel time uncertainty along the path. This involves two main challenges: 1) Predicting a path that aligns with the ground truth, and 2) modeling the impact of travel time in each segment on overall uncertainty under varying conditions. We propose DutyTTE to address these challenges. For the first challenge, we introduce a deep reinforcement learning method to improve alignment between the predicted path and the ground truth, providing more accurate travel time information from road segments to improve TTE. For the second challenge, we propose a mixture of experts guided uncertainty quantification mechanism to better capture travel time uncertainty for each segment under varying contexts. Additionally, we calibrate our results using Hoeffding's upper-confidence bound to provide statistical guarantees for the estimated confidence intervals. Extensive experiments on two real-world datasets demonstrate the superiority of our proposed method.
Paper Structure (20 sections, 9 equations, 6 figures, 6 tables)

This paper contains 20 sections, 9 equations, 6 figures, 6 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Preliminaries
  4. Methodology
  5. Path prediction from the DRL perspective
  6. Policy-gradient optimization for path prediction
  7. MoE guided uncertainty quantification
  8. Experiments
  9. Datasets
  10. Comparison Methods
  11. Overall Comparison
  12. Effectiveness of travel time uncertainty quantification
  13. Effectiveness of travel time estimation
  14. Effectiveness of path prediction
  15. Ablation Study
  16. ...and 5 more sections

Figures (6)

  • Figure 1: Motivation for DutyTTE: Each segment has varied travel time distributions, calculated using data from a period prior to the departure time.
  • Figure 2: DutyTTE first learns to predict paths from an ODT input by optimizing objectives that measure the overall alignment. It then uses travel time information from segments in the predicted path to estimate travel time confidence intervals.
  • Figure 3: Reward curves during training.
  • Figure 4: Effectiveness of hyper-parameters.
  • Figure 5: t-SNE Visualization for embeddings of road segments under varying contexts.
  • ...and 1 more figures