Research map / Reinforcement learning / Policy gradient and actor-critic
Robust Reinforcement Learning via Adversarial training with Langevin Dynamics
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use MixedNE-LD instead of GAD or Extra-Adam | DDPG · MuJoCo tasks under changed mass and friction | cumulative reward | Helped |
| Use MixedNE-LD instead of GAD or Extra-Adam | DDPG · InvertedPendulum | cumulative reward | Helped |
| Use MixedNE-LD instead of GAD or Extra-Adam | TD3 · MuJoCo tasks under mass and friction uncertainty | cumulative reward | Helped |
| Use robust TD3 with MixedNE-LD instead of SAC | TD3 · MuJoCo environments | cumulative reward | Mixed |
| Use MixedNE-LD instead of GAD or Extra-Adam | VPG · Toy MDP with varying transition parameter | cumulative reward | Helped |
| Add Langevin sampling to the actor update, using δ = 0 and K_t = 1 (ablation) | DDPG · MuJoCo tasks under changed testing conditions | cumulative reward | Helped |
| Use deterministic GAD or EG instead of MixedNE-LD | Two-player polynomial games · Non-equilibrium stationary points | distance to NE | Hurt |
| Add Langevin sampling instead of using deterministic GAD or EG | Two-player polynomial games · Initialization at a non-equilibrium stationary point | expected product θω: θ_1·ω_1 - 4η^2 (η (θ_1^2+ω_1^2) + 14η^2) < θ_1·ω_1 | Helped |
| Use MixedNE-LD rather than GAD or EG | Two-player game optimization · Objective θ²ω² with a continuum of equilibria | equilibrium discovery | No clear effect |
| Set the adversary strength to δ = 0, reducing the robust game to a one-player objective (ablation) | MixedNE-LD · MuJoCo tasks under changed testing conditions | cumulative reward | Helped |
Closest papers by meaning
- Adversarial Diffusion for Robust Reinforcement Learning (NeurIPS 2025 · similarity 0.91)
- Distributionally Robust Reinforcement Learning with Interactive Data Collection: Fundamental Hardness and Near-Optimal Algorithms (NeurIPS 2024 · similarity 0.89)
- Robust Adversarial Reinforcement Learning via Bounded Rationality Curricula (ICLR 2024 · similarity 0.89)
- Sample-Efficient Robust Multi-Agent Reinforcement Learning in the Face of Environmental Uncertainty (ICML 2024 · similarity 0.89)
- Game-Theoretic Robust Reinforcement Learning Handles Temporally-Coupled Perturbations (ICLR 2024 · similarity 0.89)
- Beyond Worst-case Attacks: Robust RL with Adaptive Defense via Non-dominated Policies (ICLR 2024 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.