Research map / Reinforcement learning / Policy gradient and actor-critic
Addressing Function Approximation Error in Actor-Critic Methods
Scott Fujimoto, Herke Hoof, David Meger
AI-extracted. The measured effects below were read by AI from the paper's full text (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 |
|---|---|---|---|
| Remove delayed policy updates from TD3 (ablation) | TD3 · HalfCheetah, Hopper, Walker2d, Ant | average return: TD3 / TD3 - DP: 9532.99 / 9590.65; 3304.75 / 2407.42; 4565.24 / 4695.50; 4185.06 / 3754.26 | Mixed |
| Remove target policy smoothing from TD3 (ablation) | TD3 · HalfCheetah, Hopper, Walker2d, Ant | average return: TD3 / TD3 - TPS: 9532.99 / 8987.69; 3304.75 / 2392.59; 4565.24 / 4033.67; 4185.06 / 4155.24 | Mixed |
| Remove Clipped Double Q-learning from TD3 (ablation) | TD3 · HalfCheetah, Hopper, Walker2d, Ant | average return: TD3 / TD3 - CDQ: 9532.99 / 9792.80; 3304.75 / 1837.32; 4565.24 / 2579.39; 4185.06 / 849.75 | Mixed |
| Add delayed policy updates to AHE (ablation) | AHE · HalfCheetah, Hopper, Walker2d, Ant | average return: AHE / AHE + DP: 8401.02 / 7588.64; 1061.77 / 1465.11; 2362.13 / 2459.53; 564.07 / 896.13 | Mixed |
| Add target policy smoothing to AHE (ablation) | AHE · HalfCheetah, Hopper, Walker2d, Ant | average return: AHE / AHE + TPS: 8401.02 / 9023.40; 1061.77 / 907.56; 2362.13 / 2961.36; 564.07 / 872.17 | Mixed |
| Add Clipped Double Q-learning to AHE (ablation) | AHE · HalfCheetah, Hopper, Walker2d, Ant | average return: AHE / AHE + CDQ: 8401.02 / 6470.20; 1061.77 / 1134.14; 2362.13 / 3979.21; 564.07 / 3818.71 | Mixed |
| Use Clipped Double Q-learning rather than the actor-critic Double Q-learning variant (ablation) | TD3 · HalfCheetah, Hopper, Walker2d, Ant | average return: TD3 / DQ-AC: 9532.99 / 9433.87; 3304.75 / 1773.71; 4565.24 / 3100.45; 4185.06 / 2445.97 | Mixed |
| Use Clipped Double Q-learning rather than the actor-critic Double DQN variant (ablation) | TD3 · HalfCheetah, Hopper, Walker2d, Ant | average return: TD3 / DDQN-AC: 9532.99 / 10306.90; 3304.75 / 2155.75; 4565.24 / 3116.81; 4185.06 / 1092.18 | Mixed |
Closest papers by meaning
- ReLU to the Rescue: Improve Your On-Policy Actor-Critic with Positive Advantages (ICML 2024 · similarity 0.89)
- Value Improved Actor Critic Algorithms (NeurIPS 2025 · similarity 0.88)
- Seizing Serendipity: Exploiting the Value of Past Success in Off-Policy Actor-Critic (ICML 2024 · similarity 0.88)
- Actor-Critics Can Achieve Optimal Sample Efficiency (ICML 2025 · similarity 0.88)
- Actor-Free Continuous Control via Structurally Maximizable Q-Functions (NeurIPS 2025 · similarity 0.88)
- Non-Asymptotic Analysis for Single-Loop (Natural) Actor-Critic with Compatible Function Approximation (ICML 2024 · similarity 0.87)
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.
