Research map / Reinforcement learning / Policy gradient and actor-critic
Gradient-Adaptive Pareto Optimization for Constrained 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 |
|---|---|---|---|
| Use GCPO instead of PPOL | PPO-based constrained RL · Goal-Lvl 1 | Ret: 13.58 -> 21.86 | Helped |
| Use GCPO instead of PPOL | PPO-based constrained RL · Button-Lvl 1 | Ret: 4.84 -> 9.38 | Helped |
| Use GCPO instead of TRPOL | PPO-based constrained RL · Button-Lvl 2 | Ret: 3.73 -> 6.44 | Helped |
| Use GCPO instead of MGDA | constrained RL · Push-Lvl 1 | Ret: 0.80 -> 2.39 | Helped |
| Use GCPO instead of PPOL | PPO-based constrained RL · Push-Lvl 2 | Ret: 1.52 -> 2.98 | Helped |
| Add gradient rebalancing (ablation) | GCPO · Goal-Lvl 1 | Ret: 13.21 -> 21.86 | Helped |
| Add gradient perturbation (ablation) | GCPO · Goal-Lvl 1 | Ret: 17.89 -> 21.86 | Helped |
| Add gradient rebalancing (ablation) | GCPO · Goal-Lvl 2 | Ret: 1.51 -> 8.39 | Helped |
| Add gradient perturbation (ablation) | GCPO · Goal-Lvl 2 | Ret: 2.23 -> 8.39 | Helped |
| Add gradient rebalancing (ablation) | GCPO · Button-Lvl 1 | Ret: 5.94 -> 9.38 | Helped |
| Add gradient perturbation (ablation) | GCPO · Button-Lvl 1 | Ret: 3.90 -> 9.38 | Helped |
| Add gradient rebalancing (ablation) | GCPO · Button-Lvl 2 | Ret: 2.18 -> 6.44 | Helped |
Closest papers by meaning
- Last-Iterate Global Convergence of Policy Gradients for Constrained Reinforcement Learning (NeurIPS 2024 · similarity 0.90)
- Conflict-Averse Gradient Aggregation for Constrained Multi-Objective Reinforcement Learning (ICLR 2025 · similarity 0.89)
- Preference Optimization on Pareto Sets: On a Theory of Multi-Objective Optimization (NeurIPS 2025 · similarity 0.89)
- Preference Conditioned Multi-Objective Reinforcement Learning: Decomposed, Diversity-Driven Policy Optimization (NeurIPS 2026 · similarity 0.88)
- PA2D-MORL: Pareto Ascent Directional Decomposition Based Multi-Objective Reinforcement Learning (AAAI 2024 · similarity 0.88)
- Reward Penalties on Augmented States for Solving Richly Constrained RL Effectively (AAAI 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.