AdaWM: Adaptive World Model based Planning for Autonomous Driving

ICLR 25 2025 50.7 method

TLDR

AdaWM adaptively finetunes world model and policy to mitigate distribution shift in autonomous driving RL.

Reasoning

The paper identifies root causes of performance degradation during finetuning and proposes adaptive mismatch identification and alignment-driven updates. Strengths include clear problem analysis and effective method; weaknesses are reliance on simulation (CARLA) and limited domain scope.

Read-first score

Read-first score 50.7, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 31.

Recency 8%

86.7

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

Methodology quality 25%

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

Topical relevance 42%

44.3

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

Reproducibility 25%

Screens links and visible text for paper, code, dataset, artifact, and repository signals. pdf=True; code=False; dataset=False; markers=none

Field roles

FrontierMethodology anchor

Rank sensitivity

Stability: volatile; rank range: 364.

Keyword Scores

world model

model-based reinforcement learning world model

world dynamics prediction

world simulator

generative world model

interactive world model

video world model

Deep Analysis

Innovations

Identification of two root causes of performance degradation in pretrain-finetune for world model based RL: mismatch of planning policy and mismatch of dynamics model due to distribution shift.
Introduction of AdaWM, an adaptive world model based planning method with mismatch identification and alignment-driven finetuning using efficient low-rank updates.

Methodology

AdaWM employs a world model based reinforcement learning framework with a pretrain-finetune paradigm. It first performs mismatch identification to quantify the mismatches between the pretrained model and the new task, then applies alignment-driven finetuning that selectively updates either the policy or the dynamics model using low-rank updates. The method is evaluated on the CARLA autonomous driving benchmark.

Key Results

Extensive experiments on CARLA driving tasks demonstrate that AdaWM significantly improves the finetuning process, leading to more robust and efficient performance in autonomous driving systems.

Resources

Paper