A Probabilistic Architecture of Long-Term Vehicle Trajectory Prediction for Autonomous Driving
DOI: 10.1016/j.eng.2021.12.020
archive: archived pipeline: cataloged verified
Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)
Summary
This paper addresses the challenge of accurate long-term vehicle trajectory prediction for autonomous vehicles (AVs) in mixed, non-connected traffic environments. Reliable prediction of surrounding vehicles is critical for safe decision-making, yet existing methods often fail to balance short-term precision with long-term uncertainty or lack interpretability. The authors propose an integrated probabilistic architecture that combines a Driving Inference Model (DIM) and a Trajectory Prediction Model (TPM) to forecast trajectories over horizons exceeding two seconds. The methodology consists of three main components. First, the DIM utilizes a dynamic Bayesian network to infer high-level driving intentions and motion characteristics. It incorporates low-level motion data (position, velocity, acceleration) and traffic rules (e.g., lane boundaries and adjacent vehicle presence) to constrain intention recognition. Model parameters are trained using the Expectation-Maximization algorithm, and probabilistic inference identifies the most likely state sequence. Second, a vehicle model-based prediction module employs a Constant Turn Rate and Acceleration (CTRA) kinematic model filtered by an Unscented Kalman Filter (UKF). This generates "support points," comprising filtered historical trajectories and short-term future predictions. Third, the TPM uses a Gaussian Process to predict long-term trajectories. It integrates the support points from the kinematic model with the intention and motion characteristics inferred by the DIM, allowing for uncertainty estimation. The approach was validated using a public naturalistic driving dataset focused on lane-changing scenarios. The authors compared their method against advanced vehicle model-based, maneuver-based, and deep learning-based approaches. The results demonstrated that the proposed architecture achieved superior performance in long-term trajectory prediction tasks. Specifically, the integration of traffic rules within the DIM improved intention recognition accuracy, while the Gaussian Process-based TPM effectively handled prediction uncertainties and provided precise long-term forecasts. The significance of this work lies in its interpretable, probabilistic framework that reduces data dependency compared to deep learning methods while addressing the limitations of purely kinematic models. By fusing high-level reasoning (intentions) with low-level physics (kinematics) and traffic constraints, the architecture provides a robust solution for situational awareness in dynamic environments. This enables AVs to make safer, more informed behavioral decisions without relying on vehicle-to-vehicle communication.
Key finding
The proposed probabilistic architecture, which integrates a dynamic Bayesian network for intention inference with a Gaussian process for trajectory prediction, outperforms existing methods in long-term vehicle trajectory prediction accuracy on naturalistic driving data.
Methodology
modeling
Provenance
The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed. Discovered via author_sweep_intake on 2026-05-28.
| Stage | Outcome | Tool | Model | Prompt | Attempts | Completed |
|---|---|---|---|---|---|---|
| discover | success | author_sweep | — | — | 2 | 2026-05-28 |
| archive | success | canonical_url | — | — | 11 | 2026-06-06 |
| extract | success | cached | — | — | 3 | 2026-06-10 |
| clean | success | clean | — | — | 1 | 2026-06-04 |
| chunk | success | chunk | — | — | 1 | 2026-06-04 |
| embed | success | embed | Qwen/Qwen3-Embedding-8B | — | 1 | 2026-06-04 |
| enrich | success | — | — | — | 1 | 2026-05-28 |
| promote | success | — | — | — | 1 | 2026-06-04 |
| summarize | success | llm | qwen3.6-27b-prismaquant | summ-v5 | 2 | 2026-06-10 |
| tag | success | vector_similarity | — | — | 15 | 2026-06-11 |
| verify | success | — | — | — | 2 | 2026-06-10 |
Summary generated by qwen3.6-27b-prismaquant on 2026-06-10; verification: verified.
Topics
Ranked by relevance to this paper. Hover a topic for its definition.
Information type
What kind of knowledge this paper contributes, grouped by family — independent of topic (what it is about) and method (how it was studied).
- Theoretical Contribution: computational model