World Models as Adversaries: Multi-Agent Self-Play Fine-Tuning for Robust Motion Planning
World Models as Adversaries: Multi-Agent Self-Play Fine-Tuning for Robust Motion Planning
This research introduces Adversarial World Modeling (AWM), a theoretically grounded framework that casts adversarially robust planner learning as a constrained min-max game. By decoupling the solver, AWM converts a planner's predictive world model into a role-conditioned adversary for inner minimization and optimizes a regret-aware robust best response for outer maximization. Experiments on nuPlan and InterPlan benchmarks demonstrate that AWM generates transferable adversarial interactions and achieves competitive closed-loop performance in both nominal and highly interactive long-tail scenarios.
The paper proposes Adversarial World Modeling (AWM), a multi-agent self-play framework for robust motion planning that converts a planner's predictive world model into an adversary to handle rare, safety-critical scenarios in dense traffic. Experiments on nuPlan and InterPlan benchmarks show competitive closed-loop performance in both nominal and long-tail scenarios.
Source: arXiv Robotics — research abstracts · Read original article ↗
Loading article text…
What the source reports
Publisher-reported claims, with original evidence. These results have not been independently verified by RoboSignal.
- paper: not_reported
- code: not_reported
- weights: not_reported
- dataset: not_reported
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
ed into a role-conditioned adversary that learns sparse, scene-adaptive attack coalitions via counterfactual credit assignment. In the outer maximization, the ego planner optimizes a regret-aware robust best response against the frozen AWM, utilizing tail-risk weighting and reference-anchored trust regions to improve hard-case recovery while preserving nominal driving behavior. Experiments on the nuPlan and InterPlan
Open source S4
arXiv:2607.10630v2 Announce Type: replace Abstract: Robust motion planning in dense traffic requires autonomous vehicles to interact in rare and safety-critical scenarios that are underrepresented in naturalistic driving data. Although adversarial training offers a feasible solution, existing methods often rely on external scenario generators, heuristic perturbations, or simulator-heavy rollouts, which makes them dif
Open source S2
benchmarks demonstrate that our method generates transferable adversarial interactions and yields a robust planner that achieves competitive closed-loop performance in both nominal and highly interactive long-tail scenarios. Theoretical analysis justifies the decoupled solver and the main optimization components.
Open source S5
ficult to integrate with modern autoregressive planners. Here, we cast adversarially robust planner learning as a constrained min-max game and propose Adversarial World Modeling (AWM), a theoretically grounded multi-agent self-play fine-tuning framework. Since solving the exact game is intractable, AWM introduces a principled decoupled solver. In the inner minimization, the planner's predictive world model is convert
Open source S3
Source：arXiv Robotics — research abstracts · arxiv.org