Problem
Research questions and friction points this paper is trying to address.
long-horizon planning
end-to-end autonomous driving
temporal dependencies
world models
safety-aware planning
GraphWorld: Long-Horizon Planning with World Models for End-to-End Autonomous Driving
🤖 AI Summary
Existing end-to-end autonomous driving approaches suffer from limited generalization and safety in complex interactive scenarios due to inadequate modeling of long-term temporal dependencies. This work proposes GraphWorld, a framework that constructs an ego-centric, adaptive neighborhood interaction graph and introduces a cross-node cross-attention mechanism to dynamically capture interactions among surrounding agents. By treating the implicit world state as a conditional signal for planning, GraphWorld enables end-to-end joint optimization of perception, prediction, and planning, facilitating long-horizon, safety-aware trajectory generation. Evaluated on Bench2Drive, NAVSIMv1/2, and nuScenes benchmarks, GraphWorld significantly reduces collision rates and substantially improves long-horizon planning performance in complex driving scenarios.
📝 Abstract
End-to-end autonomous driving has made significant progress by unifying perception, prediction, and planning within a single learning framework, achieving strong performance in short-horizon decision making. However, most existing E2E-AD methods remain confined to short-horizon planning and lack the ability to model long-term temporal dependencies, which severely limits their generalization and security in complex and highly interactive driving scenarios. In this work, we propose GraphWorld, an E2E-AD framework that explicitly enhances long-horizon planning through latent world modeling. We introduce an Ego-Centric Interaction Graph, which adaptively models critical neighboring agents based on spatial proximity, and propagates relational context to planning queries via cross-node cross-attention. We present a World-State-Conditioned Planning that learns ego-centric latent world representations by modeling interactions between an ego vehicle and surrounding agents. This latent world state captures key interaction dynamics and safety-relevant semantics, and serves as a conditioning signal to guide long-horizon, safety-aware trajectory planning. Extensive experiments on Bench2Drive, NAVSIMv1/2, and nuScenes demonstrate that GraphWorld significantly reduces collision rates and improves long-horizon planning performance, validating its effectiveness in complex driving environments.
Innovation
Methods, ideas, or system contributions that make the work stand out.
world models
long-horizon planning
ego-centric interaction graph
cross-attention
end-to-end autonomous driving
🔎 Similar Papers
No similar papers found.
💼 Related Jobs
