Research map / Reinforcement learning / Visual and representation RL
EE-RL: Vision Language Guided Reinforcement Learning with Explorer and Expert model for End-to-End Autonomous Driving
Xiaolong Li, Lan Yang, Ruyang Li, Shan Fang, Yang Liu, Xiangmo Zhao
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add multi-level attention to the explorer (ablation) | EE-RL · CARLA (train Town01, eval Town02) | DS %: 51.80 -> 81.94 | Improved |
| add LoRA fine-tuning and quantization of the VLM expert (ablation) | EE-RL · CARLA (train Town01, eval Town02) | DS %: 64.93 -> 81.94 | Improved |
| add INT4 quantization on top of attention and LoRA (ablation) | EE-RL · CARLA (train Town01, eval Town02) | DS %: 75.82 -> 81.94 | Improved |
| add INT4 quantization on top of attention and LoRA (ablation) | EE-RL · CARLA (train Town01, eval Town02) | CoR %: 4.26 -> 3.72 | No clear effect |
| add StateHash to skip redundant VLM reasoning (two VLMs) (ablation) | EE-RL · expert experiences after 1x10^6 steps | number of expert experiences: 17733 -> 90566 | Improved |
| add StateHash (single VLM) (ablation) | EE-RL · expert experiences after 1x10^6 steps | number of expert experiences: 8126 -> 53783 | Improved |
| use two VLMs in parallel instead of one (ablation) | EE-RL · expert experiences after 1x10^6 steps | number of expert experiences: 8126 -> 17733 | Improved |
| sample expert experience below 10% of batch from VLM buffer (ablation) | dual replay buffer · Town02 traffic light and obstacle avoidance | task success | Hurt |
| raise expert sample ratio to 19% (traffic light) or 24% (obstacle) (ablation) | dual replay buffer · Town02 sparse-critical scenarios | training duration: shortest at 19% and 24% | Improved |
| raise expert sample ratio beyond the optimum (ablation) | dual replay buffer · Town02 sparse-critical scenarios | training duration | Hurt |
| use TD3 or SAC backbone instead of DDPG | EE-RL · CARLA Town05-06 test | composite score (CS): 76.59 -> 78.66 (TD3), 80.09 (SAC) | Improved |
| EE-RL versus VLM-RL on collision rate | EE-RL · CARLA training Towns 01-03 | CoR %: 9.21 -> 10.76 (Town03, TD3) | Hurt |
Closest papers by meaning
- Drive-R1: Bridging Reasoning and Planning in VLMs for Autonomous Driving with Reinforcement Learning (AAAI 2026 · similarity 0.91)
- AutoVLA: A Vision-Language-Action Model for End-to-End Autonomous Driving with Adaptive Reasoning and Reinforcement Fine-Tuning (NeurIPS 2025 · similarity 0.90)
- DriveAgent-R1: Advancing VLM-based Autonomous Driving with Active Perception and Hybrid Thinking (ICLR 2026 · similarity 0.90)
- SOLVE: Synergy of Language-Vision and End-to-End Networks for Autonomous Driving (CVPR 2025 · similarity 0.89)
- $AutoDrive\text{-}P^3$: Unified Chain of Perception–Prediction–Planning Thought via Reinforcement Fine-Tuning (ICLR 2026 · similarity 0.89)
- Knowledge Distillation from Single-Task Teachers to Multi-Task Student for End-to-End Autonomous Driving (AAAI 2024 · similarity 0.89)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
