A Survey on Deep Learning for Steering Angle Prediction in Autonomous Vehicles
DOI: 10.1109/ACCESS.2020.3017883
archive: archived pipeline: cataloged verified
Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)
Summary
This survey paper addresses the critical role of steering angle prediction in the control of Autonomous Vehicles (AVs), a task essential for managing lateral motion, lane keeping, and collision avoidance. Motivated by the fact that human error accounts for over 90% of car accidents, the authors aim to provide a comprehensive review of recent advances in Deep Learning (DL) architectures applied to this specific control problem. The study is designed to assist both novice researchers by serving as an introductory resource and expert researchers by identifying gaps and open problems in the field. The authors conducted a broad study and synthesis of existing literature, creating a new comprehensive taxonomy to classify DL applications in steering angle prediction. This taxonomy organizes the field into four major categories: DL architectures, DL frameworks, AV simulators, and optimizers. The review analyzes various DL architectures, specifically detailing Convolutional Neural Networks (CNNs) and Deep Reinforcement Learning (DRL). For CNNs, the paper explains the mathematical operations of convolution, ReLU activation, and pooling layers, noting their efficiency in processing image data from front-facing cameras. For DRL, the authors outline the Markov decision processes, value functions, and Q-value functions used to train agents through trial-and-error interactions with simulated environments. The survey also examines steering control mechanisms, including the use of Ackerman’s steering geometry to relate inverse turning radius to steering angles, and discusses data preprocessing techniques such as image downsampling and normalization. The primary finding of the survey is that Convolutional Neural Networks (CNNs) are the most dominant and attractive DL architecture used for steering angle prediction, largely due to their effectiveness in image processing tasks. The authors note that while various algorithms like Genetic Algorithms and fuzzy logic have been explored, DL approaches offer better tolerance for mistakes and capability in managing unpredictable situations. However, the analysis reveals a significant limitation in the current state of research: a lack of sufficient real-world datasets. Consequently, most researchers rely heavily on data generated from simulated environments. The paper also highlights that while end-to-end models offer high accuracy, the internal workings of these networks often remain opaque, posing challenges for verification by automotive manufacturers. The significance of this work lies in its structured taxonomy and identification of open research challenges. By categorizing existing methods and highlighting the dependency on simulated data, the authors point toward promising future research directions. They propose alternative viewpoints to address the scarcity of real-world data and the need for more interpretable models. This survey serves as a foundational reference for developing novel DL approaches for AV steering control, helping to bridge the gap between theoretical algorithm development and practical, verifiable automotive applications.
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 | DOAJ | — | — | 1 | 2026-06-17 |
| archive | success | unpaywall | — | — | 1 | 2026-06-25 |
| extract | success | cached | — | — | 2 | 2026-06-25 |
| clean | success | clean | — | — | 1 | 2026-06-18 |
| chunk | success | chunk | — | — | 1 | 2026-06-18 |
| embed | success | embed | Qwen/Qwen3-Embedding-8B | — | 1 | 2026-06-18 |
| promote | success | — | — | — | 1 | 2026-06-17 |
| summarize | success | llm | qwen3.6-27b-prismaquant | summ-v5 | 1 | 2026-06-25 |
| tag | success | vector_similarity | — | — | 6 | 2026-06-18 |
| verify | success | — | — | — | 1 | 2026-06-26 |
Summary generated by qwen3.6-27b-prismaquant on 2026-06-25; 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