⌁ Awesome World Model Hub Papers · Datasets · Projects
中文 GitHub
← Back to papers

From Imitation to Exploration: End-to-end Autonomous Driving based on World Model

arXiv 25.4 2025 71 method, application

TLDR

Proposes RAMBLE, an end-to-end world model-based RL method for autonomous driving combining imitation learning and reinforcement learning.

Reasoning

The paper presents a novel integration of IL and RL via a world model, achieving strong simulation results. However, it lacks real-world validation and provides limited architectural details.

Read-first score

Read-first score 71, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 42.

Recency 8%
86.7

Uses a gentle age decay so recent papers surface without erasing older foundations. 2025

Reproducibility 25%
85

Screens links and visible text for paper, code, dataset, artifact, and repository signals. pdf=True; code=True; dataset=False; markers=code,dataset,github

Methodology quality 25%
70

Screens visible abstract and analysis fields for experiment, dataset, baseline, metric, and limitation evidence. markers=dataset,metric

Topical relevance 42%
60

Uses existing LLM keyword relevance scores normalized to 0-100. world model,world simulator,generative world model,interactive world model,video world model,world dynamics prediction,model-based reinforcement learning world model

Field roles

FrontierMethodology anchorReproducibility anchor

Rank sensitivity

Stability: volatile; rank range: 199.

Keyword Scores

world model
9
model-based reinforcement learning world model
9
world dynamics prediction
8
interactive world model
6
generative world model
4
world simulator
3
video world model
3

Deep Analysis

Innovations

  • Combining imitation learning and reinforcement learning with a training scheme that initializes the policy network using IL and employs KL loss and soft update mechanisms to smoothly transition from IL to RL
  • Asymmetrical variational autoencoder for extracting environmental context from RGB images and LiDAR data
  • Transformer-based architecture to capture dynamic transitions of traffic participants
  • End-to-end world model-based reinforcement learning method for driving decision-making

Methodology

RAMBLE uses an asymmetrical variational autoencoder to extract environmental context from RGB images and LiDAR data, then a transformer-based architecture captures dynamic transitions of traffic participants. An actor-critic reinforcement learning algorithm derives driving strategies based on latent features of the current state and dynamics. The policy network is initialized using imitation learning, and KL loss with soft update mechanisms are employed to smoothly transition the model from IL to RL.

Key Results

RAMBLE achieves state-of-the-art performance in route completion rate on the CARLA Leaderboard 1.0 and completes all 38 scenarios on the CARLA Leaderboard 2.0.

Limitations

  • The method is evaluated only in the CARLA simulator, and its performance in real-world driving scenarios is not demonstrated
  • Potential computational complexity due to the use of variational autoencoder, transformer, and actor-critic components

Resources

PaperCode

Tags