Vehicles Trajectory Prediction Using Recurrent VAE Network
DOI: 10.1109/access.2022.3161661
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 vehicle trajectory prediction for autonomous driving, specifically focusing on modeling the uncertainty and non-linear behavior of surrounding vehicles in highway scenarios. The authors propose a deep learning architecture combining Recurrent Neural Networks (LSTM) and Variational Auto-Encoders (VAE) to predict future local trajectories and maneuvers. The motivation stems from the limitations of existing physics-based and maneuver-based models, which often fail to capture complex vehicle interactions or generate realistic, naturalistic driving paths. The proposed method leverages the generative properties of VAEs to encode similar driving situations into a structured latent space and decode them into realistic trajectories, while also providing uncertainty estimates based on input data quality. The study utilizes the highD dataset, a high-quality highway driving dataset, split into training (75%) and testing (25%) sequences. The model architecture consists of two separate VAEs: one ($VAE_{surr}$) encodes surrounding vehicle data (distances and speeds) into a 7-dimensional latent space, and the other ($VAE_{xy}$) encodes the ego-vehicle’s past trajectory into a 2-dimensional latent space. These latent representations are fed into a prediction model composed of fully connected layers, which predicts the future latent state. This state is then decoded by $VAE_{xy}$ to generate the future trajectory. The model is trained in two stages: first, the VAEs are trained for reconstruction, and second, the prediction module is trained while keeping the VAE weights fixed. The authors also perform a theoretical analysis of the latent space, using Principal Component Analysis to determine dimensionality and visualizing the latent space to confirm that similar maneuvers cluster together. The results demonstrate that the proposed Recurrent VAE network outperforms state-of-the-art baseline methods, including 3D CNN-LSTM and Multi-Head Attention models, in terms of Mean Absolute Error (MAE) and Root Mean Square Error (RMSE), particularly for longer prediction horizons (up to 5 seconds). The analysis of the latent space confirms that the VAE encoder successfully distinguishes between different maneuvers (lane keeping vs. lane changing) and generates a smooth, continuous representation of trajectories. Furthermore, the model’s ability to estimate uncertainty is validated through experiments with noisy input data; the variance of the predicted trajectory increases significantly when input data quality degrades, whereas it remains low for clear, known situations. This indicates that the model can reliably quantify its confidence in predictions. The significance of this work lies in its integration of generative modeling with trajectory prediction, offering a robust solution that not only improves prediction accuracy but also provides crucial uncertainty estimates for safe autonomous navigation. By proving that VAEs can create a structured latent space where similar driving contexts are proximal, the paper highlights the potential of probabilistic deep learning models to handle the unpredictability of real-world traffic. The approach is computationally efficient enough for real-time applications and does not rely on explicit maneuver classifiers, making it flexible and generalizable to various highway scenarios.
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.
| Stage | Outcome | Tool | Model | Prompt | Attempts | Completed |
|---|---|---|---|---|---|---|
| discover | success | Crossref | — | — | 1 | 2026-06-25 |
| archive | success | unpaywall | — | — | 2 | 2026-06-26 |
| extract | success | cached | — | — | 2 | 2026-06-26 |
| clean | success | clean | — | — | 1 | 2026-06-26 |
| chunk | success | chunk | — | — | 1 | 2026-06-26 |
| embed | success | embed | Qwen/Qwen3-Embedding-8B | — | 1 | 2026-06-26 |
| enrich | success | openalex | — | — | 1 | 2026-06-26 |
| promote | success | — | — | — | 1 | 2026-06-25 |
| summarize | success | llm | qwen3.6-27b-prismaquant | summ-v5 | 1 | 2026-06-26 |
| tag | success | vector_similarity | — | — | 6 | 2026-06-26 |
| verify | success | — | — | — | 1 | 2026-06-26 |
Summary generated by qwen3.6-27b-prismaquant on 2026-06-26; 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