Research map / Reinforcement learning / Offline RL
Risk-Averse Offline Reinforcement Learning
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 |
|---|---|---|---|
| optimize CVaR using a distributional critic instead of the risk-neutral objective | RAAC versus D4PG · Car | CVaR₀.₁: 15.6 (4.4) -> 48.0 (8.3) | Helped |
| use a full offline method with VAE imitation regularization rather than RAAC without the VAE (ablation) | O-RAAC · HalfCheetah-medium | CVaR₀.₁: -55 (1) -> 214 (36) | Helped |
| use a full offline method with VAE imitation regularization rather than RAAC without the VAE (ablation) | O-RAAC · HalfCheetah-expert | CVaR₀.₁: 3 (13) -> 595 (191) | Helped |
| use a full offline method with VAE imitation regularization rather than RAAC without the VAE (ablation) | O-RAAC · Walker2D-medium | CVaR₀.₁: 55 (2) -> 751 (154) | Helped |
| use a full offline method with VAE imitation regularization rather than RAAC without the VAE (ablation) | O-RAAC · Walker2D-expert | CVaR₀.₁: 54 (2) -> 1172 (71) | Helped |
| use a full offline method with VAE imitation regularization rather than RAAC without the VAE (ablation) | O-RAAC · Hopper-medium | CVaR₀.₁: 71 (23) -> 1416 (28) | Helped |
| use a full offline method with VAE imitation regularization rather than RAAC without the VAE (ablation) | O-RAAC · Hopper-expert | CVaR₀.₁: 474 (0) -> 980 (28) | Helped |
| use risk-averse training rather than risk-neutral O-D4PG | O-RAAC versus O-D4PG · HalfCheetah-medium | CVaR₀.₁: 66 (34) -> 214 (36) | Helped |
| use risk-averse training rather than risk-neutral O-D4PG | O-RAAC versus O-D4PG · HalfCheetah-expert | CVaR₀.₁: 556 (263) -> 595 (191) | Helped |
| use risk-averse training rather than risk-neutral training | O-RAAC versus risk-neutral benchmarks · Offline MuJoCo tasks | CVaR₀.₁ | Helped |
| use O-RAAC with VAE imitation rather than the VAE as a pure behavior-policy imitator (ablation) | O-RAAC · Offline MuJoCo tasks | CVaR₀.₁ | Helped |
| vary the perturbation strength λ toward zero, making the policy more purely imitative (ablation) | O-RAAC · Offline MuJoCo tasks | CVaR₀.₁ | Hurt |
Closest papers by meaning
- Conservative Offline Distributional Reinforcement Learning (NeurIPS 2021 · similarity 0.92)
- RAMAC: Multimodal Risk-Aware Offline Reinforcement Learning and the Role of Behavior Regularization (ICML 2026 · similarity 0.90)
- Adversarially Trained Weighted Actor-Critic for Safe Offline Reinforcement Learning (NeurIPS 2024 · similarity 0.89)
- Offline-Boosted Actor-Critic: Adaptively Blending Optimal Historical Behaviors in Deep Off-Policy RL (ICML 2024 · similarity 0.89)
- Online Optimization for Offline Safe Reinforcement Learning (NeurIPS 2025 · similarity 0.89)
- Risk-Conditioned Reinforcement Learning: A Generalized Approach for Adapting to Varying Risk Measures (AAAI 2024 · similarity 0.89)
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.