Research map / Reinforcement learning / Policy gradient and actor-critic
Softmax Deep Double Deterministic Policy Gradients
Ling Pan, Qingpeng Cai, Longbo Huang
Read the paperCheck your idea against 137,217 papers
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 |
|---|---|---|---|
| Use the softmax operator with double critics | TD3 · MuJoCo and Box2d continuous-control tasks | performance | Improved |
| Use the softmax operator with a single critic | DDPG · Hopper-v2 and Walker2d-v2 | performance | Improved |
| Use the softmax operator with double critics | TD3 · MuJoCo environments | absolute value-estimation bias | Improved |
| Use a deterministic SAC variant with log-sum-exp | DDPG · Hopper-v2 and Walker2d-v2 | performance | Hurt |
| Use a deterministic SAC variant with log-sum-exp | DDPG · Hopper-v2 and Walker2d-v2 | absolute value-estimation bias | Hurt |
| Use SD3 instead of SAC | SAC · Continuous-control environments except Humanoid-v2 | performance | Improved |
| Add a second actor (ablation) | SD3 · HalfCheetah-v2 | performance | Improved |
| Add a second actor (ablation) | TD3 · HalfCheetah-v2 | performance | Improved |
| Set the number of action-noise samples to 2 (ablation) | SD3 · HalfCheetah-v2 | sample efficiency | Hurt |
| Use multiple sampled actions before taking the minimum, without the softmax operator (ablation) | TD3 · HalfCheetah-v2 | performance | Mixed |
| Average the sampled Q-values instead of applying the softmax operator (ablation) | SD3 · HalfCheetah-v2 | performance | Hurt |
| Directly apply the softmax estimate and then clip it by the original critic value, as in TD3 (ablation) | TD3 · Continuous-control tasks | underestimation bias | Hurt |
Closest papers by meaning
- Langevin Soft Actor-Critic: Efficient Exploration through Uncertainty-Driven Critic Learning (ICLR 2025 · similarity 0.88)
- Generative Actor-Critic with Soft Bridge Policies (NeurIPS 2026 · similarity 0.87)
- Mean Field Langevin Actor-Critic: Faster Convergence and Global Optimality beyond Lazy Learning (ICML 2024 · similarity 0.86)
- ReLU to the Rescue: Improve Your On-Policy Actor-Critic with Positive Advantages (ICML 2024 · similarity 0.86)
- Value Improved Actor Critic Algorithms (NeurIPS 2025 · similarity 0.86)
- Actor-Free Continuous Control via Structurally Maximizable Q-Functions (NeurIPS 2025 · similarity 0.86)
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.
