High-Speed Cornering Control and Real-Vehicle Deployment for Autonomous Electric Vehicles
DOI: 10.48550/arxiv.2411.11762
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 deploying reinforcement learning (RL) controllers for high-speed drift cornering in autonomous electric vehicles, specifically focusing on bridging the simulation-to-reality gap. While RL has shown promise in optimizing cornering times in simulated environments, discrepancies between simulated dynamics and real-world conditions—such as sensor noise, actuator constraints, and environmental variability—have hindered practical deployment. The authors propose a novel control framework that integrates trajectory optimization with a hybrid RL-Model Predictive Control (MPC) mechanism to enable safe and efficient drift maneuvers on consumer-grade electric vehicles. The methodology consists of two primary phases: controller design and real-vehicle deployment. For controller design, the authors employ a Twin Delayed Deep Deterministic Policy Gradient (TD3) algorithm trained in a CarSim simulation environment. To enhance training efficiency and reward density, they implement a pre-trajectory optimization strategy using Bézier curves based on the principle of minimum curvature. This generates smooth reference paths and speed plans that serve as immediate rewards during the RL training process, guiding the agent toward optimal cornering times while penalizing deviations and unsafe states. The reward function combines immediate rewards for trajectory tracking and high side-slip angles with terminal rewards for minimizing total cornering time. For real-world deployment, the authors introduce an RL-MPC fusion mechanism. The TD3-derived optimal trajectory serves as the primary input, while a two-step forward model predictive controller provides corrective inputs to track this trajectory precisely, compensating for simulation-to-reality discrepancies. The study validates the proposed framework through both simulation and real-vehicle tests. In simulation, the Bézier-based pre-trajectory optimization demonstrated superior performance compared to using road centerlines, reducing cornering times for 90-degree, 135-degree, and U-turn scenarios by approximately 4–9%. The TD3 algorithm successfully learned transient drift maneuvers that minimized cornering time. Crucially, the authors conducted real-vehicle tests on consumer-grade electric vehicles, executing U-turns and right-angle drift turns. The hybrid RL-MPC controller enabled the vehicle to perform these extreme maneuvers safely and effectively, demonstrating robust trajectory tracking despite the nonlinear dynamics and unpredictable conditions of the real world. The significance of this work lies in being the first to deploy a scenario-to-action RL-based transient drift cornering algorithm on consumer-grade electric vehicles. By integrating model-based MPC with data-driven RL, the authors provide a viable solution for bridging the simulation-to-reality gap, enhancing the adaptability of RL controllers in unpredictable environments. This approach not only advances autonomous racing by reducing lap times but also improves the handling capabilities of high-performance autonomous vehicles during extreme driving conditions, contributing to the broader field of autonomous vehicle control and safety.
Key finding
A hybrid RL-MPC fusion control framework successfully enables the deployment of RL-based transient drift cornering algorithms on consumer-grade electric vehicles, achieving stable high-speed maneuvers by using MPC to correct for simulation-to-reality discrepancies.
Methodology
on_road
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 | 2026-06-04 |
| 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