A survey of deep learning techniques for autonomous driving
DOI: 10.1002/rob.21918
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 reviews the state-of-the-art deep learning and artificial intelligence technologies applied to autonomous driving, motivated by the rapid progress in self-driving vehicle technology over the last decade. The authors aim to provide a comprehensive overview of methodologies used in driving scene perception, path planning, behavior arbitration, and motion control. The study addresses the transition from classical non-learning approaches to AI-based systems, examining both modular perception-planning-action pipelines and End-to-End systems that directly map sensory information to steering commands. Additionally, the paper investigates critical challenges in designing AI architectures for autonomous vehicles, including safety, training data sources, and computational hardware requirements. The authors structure their review around three primary deep learning paradigms: Convolutional Neural Networks (CNNs), Recurrent Neural Networks (RNNs), and Deep Reinforcement Learning (DRL). CNNs are analyzed as spatial feature extractors that mimic the mammalian visual cortex, utilizing convolutional filters to learn translation-invariant features from image data. RNNs, specifically Long Short-Term Memory (LSTM) networks, are examined for their ability to process temporal sequence data by solving the vanishing gradient problem through gating mechanisms. DRL is framed within the Partially Observable Markov Decision Process (POMDP) formalism, where agents learn optimal driving policies by maximizing cumulative discounted rewards. The survey details specific DRL advancements, including Deep Q-Networks (DQN) and the Rainbow algorithm, which incorporates extensions like Double Q Learning and Prioritized Replay to improve performance. The findings categorize deep learning applications into hierarchical components: perception and localization, high-level path planning, behavior arbitration, and motion controllers. The paper highlights that while modular pipelines allow for hybrid designs combining AI and classical methods, End-to-End systems offer a unified approach to mapping observations to control outputs. The review identifies that CNNs are dominant in object detection and semantic segmentation, while RNNs handle temporal dependencies in video streams. For decision-making, DRL provides a framework for learning navigation policies, though it requires significant simulation and modeling of vehicle behavior. The authors also note that current implementations often rely on simulated environments due to the complexity of real-world state estimation from sensor readings. The significance of this work lies in its systematic comparison of deep learning strengths and limitations in autonomous driving, assisting researchers and engineers in design choices. By covering the full spectrum from sensory hardware debates (camera vs. LiDAR) to deployment constraints, the survey provides a foundational reference for understanding how AI integrates into the autonomous driving stack. It underscores the necessity of addressing safety monitors and robust training data sources to ensure reliable performance. The paper concludes that while deep learning has become central to autonomous driving, particularly post-2005, challenges remain in ensuring safety and managing the computational demands of these complex architectures.
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 | OpenAlex-citations | — | — | 1 | 2026-06-19 |
| archive | success | unpaywall | — | — | 2 | 2026-06-26 |
| extract | success | cached | — | — | 2 | 2026-06-26 |
| clean | success | clean | — | — | 1 | 2026-06-20 |
| chunk | success | chunk | — | — | 1 | 2026-06-20 |
| embed | success | embed | Qwen/Qwen3-Embedding-8B | — | 1 | 2026-06-20 |
| promote | success | — | — | — | 1 | 2026-06-19 |
| summarize | success | llm | qwen3.6-27b-prismaquant | summ-v5 | 1 | 2026-06-26 |
| tag | success | vector_similarity | — | — | 6 | 2026-06-20 |
| 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