mlsanityData to 7 Oct 2026

RAD: Training an End-to-End Driving Policy via Large-Scale 3DGS-based Reinforcement Learning

arXiv 2502.13144 ↗First posted 18 Feb 2025cs.CV cs.ROcited by 76

On the map: End-to-end autonomous driving → Imitation-learning end-to-end planners

also in 1 more
Abstract (arXiv)

Existing end-to-end autonomous driving (AD) algorithms typically follow the Imitation Learning (IL) paradigm, which faces challenges such as causal confusion and an open-loop gap. In this work, we propose RAD, a 3DGS-based closed-loop Reinforcement Learning (RL) framework for end-to-end Autonomous Driving. By leveraging 3DGS techniques, we construct a photorealistic digital replica of the real physical world, enabling the AD policy to extensively explore the state space and learn to handle out-of-distribution scenarios through large-scale trial and error. To enhance safety, we design specialized rewards to guide the policy in effectively responding to safety-critical events and understanding real-world causal relationships. To better align with human driving behavior, we incorporate IL into RL training as a regularization term. We introduce a closed-loop evaluation benchmark consisting of diverse, previously unseen 3DGS environments. Compared to IL-based methods, RAD achieves stronger performance in most closed-loop metrics, particularly exhibiting a 3x lower collision rate. Abundant closed-loop results are presented in the supplementary material. Code is available at https://github.com/hustvl/RAD for facilitating future research.

Key facts. Cited by 76 papers in the map (data through 7 Oct 2026). Placed in Imitation-learning end-to-end planners, one of 9 directions in End-to-end autonomous driving. Builds on 41 of its 53 references that are in the map. First posted on arXiv 18 Feb 2025.

Builds on

41 of 53 references.

Closest ancestors · 5

Same direction, newest first

  1. 2024DiffusionDrive: Truncated Diffusion Model for End-to-End Autonomous Driving22 Nov
  2. 2024VADv2: End-to-End Vectorized Autonomous Driving via Probabilistic Planning20 Feb
  3. 2023PPAD: Iterative Interactions of Prediction and Planning for End-to-end Autonomous Driving14 Nov
  4. 2022Imitation Is Not Enough: Robustifying Imitation with Reinforcement Learning for Challenging Driving Scenarios21 Dec
  5. 2021End-to-End Urban Driving by Imitating a Reinforcement Learning Coach18 Aug

Other references · 36

  1. 2015Deep Residual Learning for Image Recognitioncited by 51,809
  2. 2020An Image is Worth 16x16 Words: Transformers for Image Recognition at Scalecited by 29,915
  3. 2017Proximal Policy Optimization Algorithmscited by 14,441
  4. 2025DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learningcited by 6,146
  5. 20233D Gaussian Splatting for Real-Time Radiance Field Renderingcited by 6,141
  6. 2017CARLA: An Open Urban Driving Simulatorcited by 2,431
  7. 2020Shortcut Learning in Deep Neural Networkscited by 1,928
  8. 2015High-Dimensional Continuous Control Using Generalized Advantage Estimationcited by 1,916
  9. 2022BEVFormer: Learning Bird's-Eye-View Representation from Multi-Camera Images via Spatiotemporal Transformerscited by 1,213
  10. 2022Planning-oriented Autonomous Drivingcited by 1,068
  11. 2023VAD: Vectorized Scene Representation for Efficient Autonomous Drivingcited by 613
  12. 2021Multi-Modal Fusion Transformer for End-to-End Autonomous Drivingcited by 372
24 more
  1. 2023DrivingGaussian: Composite Gaussian Splatting for Surrounding Dynamic Autonomous Driving Scenescited by 359
  2. 2022MapTR: Structured Modeling and Learning for Online Vectorized HD Map Constructioncited by 311
  3. 2023Driving into the Future: Multiview Visual Forecasting and Planning with World Model for Autonomous Drivingcited by 309
  4. 2023UniSim: A Neural Closed-Loop Sensor Simulatorcited by 294
  5. 2024Hydra-MDP: End-to-end Multimodal Planning with Multi-target Hydra-Distillationcited by 244
  6. 2023Is Ego Status All You Need for Open-Loop End-to-End Autonomous Driving?cited by 242
  7. 2024SparseDrive: End-to-End Autonomous Driving via Sparse Scene Representationcited by 239
  8. 2024GenAD: Generative End-to-End Autonomous Drivingcited by 225
  9. 2024Street Gaussians: Modeling Dynamic Urban Scenes with Gaussian Splattingcited by 218
  10. 2023MapTRv2: An End-to-End Framework for Online Vectorized HD Map Constructioncited by 192
  11. 2023Rethinking the Open-Loop Evaluation of End-to-End Autonomous Driving in nuScenescited by 191
  12. 2023MARS: An Instance-aware, Modular and Realistic Simulator for Autonomous Drivingcited by 170
  13. 2024Enhancing End-to-End Autonomous Driving with Latent World Modelcited by 161
  14. 2019End-to-End Model-Free Reinforcement Learning for Urban Driving using Implicit Affordancescited by 147
  15. 2023NeuRAD: Neural Rendering for Autonomous Drivingcited by 146
  16. 2021Learning to drive from a world on railscited by 114
  17. 2018CIRL: Controllable Imitative Reinforcement Learning for Vision-based Self-drivingcited by 96
  18. 2024HUGSIM: A Real-Time, Photo-Realistic and Closed-Loop Simulator for Autonomous Drivingcited by 84
  19. 2023Lane Graph as Path: Continuity-preserving Path-wise Modeling for Online Lane Graph Constructioncited by 50
  20. 2021Efficient Deep Reinforcement Learning with Imitative Expert Priors for Autonomous Drivingcited by 46
  21. 2024Solving Motion Planning Tasks with a Scalable Generative Modelcited by 40
  22. 2024Producing and Leveraging Online Map Uncertainty in Trajectory Predictioncited by 32
  23. 2022Perceive, Interact, Predict: Learning Dynamic and Static Clues for End-to-End Motion Predictioncited by 28
  24. 2022CADRE: A Cascade Deep Reinforcement Learning Framework for Vision-based Autonomous Urban Drivingcited by 16

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2025: 37; 2026: 39.

76 papers.

Sort:

By area:

  1. 2025AutoVLA: A Vision-Language-Action Model for End-to-End Autonomous Driving with Adaptive Reasoning and Reinforcement Fine-Tuningcited by 252
  2. 2025ReCogDrive: A Reinforced Cognitive Framework for End-to-End Autonomous Drivingcited by 190
  3. 2025End-to-End Driving with Online Trajectory Evaluation via BEV World Modelcited by 145
  4. 2025mimic-video: Video-Action Models for Generalizable Robot Control Beyond VLAscited by 133
  5. 2025AlphaDrive: Unleashing the Power of VLMs in Autonomous Driving via Reinforcement Learning and Reasoningcited by 108
  6. 2025DiffusionDriveV2: Reinforcement Learning-Constrained Truncated Diffusion Modeling in End-to-End Autonomous Drivingcited by 52
  7. 2025DriveLaW:Unifying Planning and Video Generation in a Latent Driving Worldcited by 43
  8. 2025Centaur: Robust End-to-End Autonomous Driving with Test-Time Trainingcited by 41
  9. 2025How Far are VLMs from Visual Spatial Intelligence? A Benchmark-Driven Perspectivecited by 41
  10. 2025RAP: 3D Rasterization Augmented End-to-End Planningcited by 39
  11. 2025SimScale: Learning to Drive via Real-World Simulation at Scalecited by 34
  12. 2025OmniNWM: Omniscient Driving Navigation World Modelscited by 32
38 more
  1. 2025Vision-Language-Action Models for Autonomous Driving: Past, Present, and Futurecited by 32
  2. 2025IRL-VLA: Training an Vision-Language-Action Policy via Reward World Modelcited by 31
  3. 2025MindDrive: A Vision-Language-Action Model for Autonomous Driving via Online Reinforcement Learningcited by 25
  4. 2025Reinforced Refinement with Self-Aware Expansion for End-to-End Autonomous Drivingcited by 19
  5. 2025MTGS: Multi-Traversal Gaussian Splattingcited by 17
  6. 2025DIVER: Reinforced Diffusion Breaks Imitation Bottlenecks in End-to-End Autonomous Drivingcited by 17
  7. 2025PRIX: Learning to Plan from Raw Pixels for End-to-End Autonomous Drivingcited by 16
  8. 2025Zero-Human Demonstration End-to-end Autonomous Driving with Trajectory Scorercited by 16
  9. 2025AD-R1: Closed-Loop Reinforcement Learning for End-to-End Autonomous Driving with Impartial World Modelscited by 13
  10. 2026DreamerAD: Efficient Reinforcement Learning via Latent World Model for Autonomous Drivingcited by 13
  11. 2025ResAD: Normalized Residual Trajectory Modeling for End-to-End Autonomous Drivingcited by 12
  12. 2025HAIF-GS: Hierarchical and Induced Flow-Guided Gaussian Splatting for Dynamic Scenecited by 11
  13. 2026MindDriver: Introducing Progressive Multimodal Reasoning for Autonomous Drivingcited by 10
  14. 2026CoWorld-VLA: Thinking in a Multi-Expert World Model for Autonomous Drivingcited by 10
  15. 2025Learning-based 3D Reconstruction in Autonomous Driving: A Comprehensive Surveycited by 9
  16. 2026Uni-World VLA: Interleaved World Modeling and Planning for Autonomous Drivingcited by 9
  17. 2026BridgeSim: Unveiling the OL-CL Gap in End-to-End Autonomous Drivingcited by 9
  18. 2025TakeAD: Preference-based Post-optimization for End-to-end Autonomous Driving with Expert Takeover Datacited by 8
  19. 2025Generative AI for Autonomous Driving: Frontiers and Opportunitiescited by 7
  20. 2025Model-Based Policy Adaptation for Closed-Loop End-to-End Autonomous Drivingcited by 7
  21. 2026ParkGaussian: Surround-view 3D Gaussian Splatting for Autonomous Parkingcited by 7
  22. 2026RAD-2: Scaling Reinforcement Learning in a Generator-Discriminator Frameworkcited by 7
  23. 2025From Coarse to Fine: Learnable Discrete Wavelet Transforms for Efficient 3D Gaussian Splattingcited by 6
  24. 2026Generative Scenario Rollouts for End-to-End Autonomous Drivingcited by 6
  25. 2026Towards Interactive Video World Modeling: Frontiers, Challenges, Benchmarks, and Future Trendscited by 6
  26. 2026World Engine: Towards the Era of Post-Training for Autonomous Drivingcited by 6
  27. 2025B'ezierGS: Dynamic Urban Scene Reconstruction with B'ezier Curve Gaussian Splattingcited by 5
  28. 2025VDRive: Leveraging Reinforced VLA and Diffusion Policy for End-to-end Autonomous Drivingcited by 5
  29. 2025Prioritizing Perception-Guided Self-Supervision: A New Paradigm for Causal Modeling in End-to-End Autonomous Drivingcited by 5
  30. 2026Xiaomi Auto World Model: A Joint World Model Integrating Reconstruction and Generation for Autonomous Drivingcited by 5
  31. 2025WPT: World-to-Policy Transfer via Online World Model Distillationcited by 4
  32. 2026Senna-2: Aligning VLM and End-to-End Driving Policy for Consistent Decision Making and Planningcited by 4
  33. 2026Fine-tuning is Not Enough: A Parallel Framework for Collaborative Imitation and Reinforcement Learning in End-to-end Autonomous Drivingcited by 4
  34. 2026HAD: Combining Hierarchical Diffusion with Metric-Decoupled RL for End-to-End Drivingcited by 4
  35. 2026MDrive: Benchmarking Closed-Loop Cooperative Driving for End-to-End Multi-agent Systemscited by 4
  36. 2026Scaling Self-Play for End-to-End Drivingcited by 4
  37. 2026UFO: Unifying Feed-Forward and Optimization-based Methods for Large Driving Scene Modelingcited by 3
  38. 2026EvoDriveVLA: Evolving Driving VLA Models via Collaborative Perception-Planning Distillationcited by 3

50 most cited of 76 All 76 (JSON)

Similar, not cited

Close in wording, no citation either way

  1. 2024From Imitation to Exploration: End-to-end Autonomous Driving based on World Model
  2. 2024Efficient and Generalized end-to-end Autonomous Driving System with Latent Deep Reinforcement Learning and Demonstrations
  3. 2025V-Max: A Reinforcement Learning Framework for Autonomous Driving
  4. 2024Autonomous Algorithm for Training Autonomous Vehicles with Minimal Human Intervention
  5. 2026Post-Training in End-to-End Autonomous Driving
  6. 2020Learning to Drive (L2D) as a Low-Cost Benchmark for Real-World Reinforcement Learning
  7. 2026Planning-Oriented End-to-End Autonomous Driving: Architectures, Evaluation, and Emerging Paradigms
  8. 2025A Comprehensive Review of Reinforcement Learning for Autonomous Driving in the CARLA Simulator