Online Physical Enhanced Residual Learning for Connected Autonomous Vehicles Platoon Centralized Control
DOI: 10.1109/iv55156.2024.10588534
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 maintaining stability and safety in Connected Autonomous Vehicle (CAV) platoons operating in dynamic, unpredictable traffic environments. Existing control methods face significant limitations: physics-based approaches, such as Model Predictive Control (MPC), offer interpretability and safety constraints but often oversimplify complex dynamics and struggle with real-world disturbances. Conversely, data-driven machine learning methods, like Reinforcement Learning (RL), can handle non-linear relationships but lack transparency and require extensive pre-training data, which is risky to collect. To bridge this gap, the authors propose an Online Physical Enhanced Residual Learning (PERL) framework that synergistically combines a centralized MPC controller with online Q-learning. The methodology integrates two modules: a fundamental physical-based controller and a learning-based residual controller. The centralized MPC component uses vehicle velocity as the control input to optimize multi-objective cooperative goals, including desired acceleration, velocity, and inter-vehicle distances, while adhering to safety constraints. This physical model provides prior knowledge and ensures theoretical stability. The residual learning module employs online Q-learning to correct errors caused by model uncertainties, calibration inaccuracies, and external disturbances. To handle continuous state and action spaces, the study discretizes control outputs and applies fuzzy logic to categorize continuous states into discrete groups. The Q-learning agent updates its policy every 20 time steps based on the discrepancy between desired and actual speeds, thereby refining the MPC’s output in real-time. The framework was evaluated in a simulation environment involving a platoon of five vehicles over 15-second durations. Two scenarios were tested: uniform motion at 15 m/s and variable speed involving acceleration and deceleration phases. The study introduced affine and quadratic errors to simulate real-world control transfer inaccuracies. The PERL controller using Q-learning was compared against a standalone MPC controller and a PERL variant using a neural network for residual learning. Results demonstrated that the Q-learning-based PERL significantly outperformed the other methods. Specifically, the cumulative absolute position and velocity errors were, on average, 86.73% and 55.28% lower than those of the standalone MPC controller. Furthermore, the Q-learning approach reduced cumulative position and velocity errors by 18.83% and 12.82%, respectively, compared to the neural network-based PERL. In complex variable-speed scenarios, the maximum absolute error gaps between the Q-learning PERL and MPC alone reached 88.33% for position and 71.24% for velocity. The significance of this work lies in its ability to combine the interpretability and safety guarantees of physics-based models with the adaptive accuracy of data-driven learning. The PERL framework maintains the transparency of MPC while significantly improving computational efficiency and control precision under diverse conditions. By correcting residuals online, the system achieves rapid convergence and superior stability, offering a robust solution for safe and precise motion control in autonomous vehicle platoons facing real-world uncertainties.
Key finding
The online physical enhanced residual learning framework using Q-learning significantly outperforms model predictive control alone and neural network-based residual learning in reducing position and velocity errors for connected autonomous vehicle platoons.
Methodology
simulation_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 | — | — | 1 | 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.