Research map / Reinforcement learning / Policy gradient and actor-critic

Preference-Guided Adversarial Policy Optimization for Long-Tail Robust Driving

NeurIPS 2026 · Reinforcement learning › Policy gradient and actor-critic · approach: Safe policy improvement

Read the paperCheck your idea against 108,967 papers

AI-extracted summary. Written by AI from the paper's abstract only. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Proposes a preference-guided adversarial policy optimization framework that jointly trains a scenario generator and driving policy for long-tail robust autonomous driving.

Problem

Autonomous driving policies need robustness to rare, safety-critical long-tail scenarios; existing adversarial training decouples scenario generation from policy optimization and relies on heuristic surrogate objectives, causing objective misalignment and failure to track evolving policy failure mod

Approach

Couples adversarial scenario generation with policy optimization using preference-guided signals instead of heuristic surrogates, so the adversary adapts to the shifting failure modes of the evolving policy.

Closest papers by meaning

• World Models as Adversaries: Multi-Agent Self-Play Fine-Tuning for Robust Motion Planning (NeurIPS 2026 · similarity 0.90)
• Robust Multi-Agent Reinforcement Learning with Stochastic Adversary (ICML 2025 · similarity 0.87)
• Robust Adversarial Reinforcement Learning via Bounded Rationality Curricula (ICLR 2024 · similarity 0.87)
• An Optimistic Algorithm for online CMDPS with Anytime Adversarial Constraints (ICML 2025 · similarity 0.86)
• Beyond Worst-case Attacks: Robust RL with Adaptive Defense via Non-dominated Policies (ICLR 2024 · similarity 0.86)
• Robust and Diverse Multi-Agent Learning via Rational Policy Gradient (NeurIPS 2025 · similarity 0.86)