Research map / Reinforcement learning / Visual and representation RL
Structured Control Nets for Deep 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 |
|---|---|---|---|
| Add an additive linear control module to the nonlinear policy | SCN policy · MuJoCo and Roboschool | training average episodic reward: 13.2% for PPO; 17.5% for ES | Helped |
| Use SCN-16 instead of MLP-64 | ES policy on six MuJoCo environments · MuJoCo | network size: 15.6% | Helped |
| Remove the nonlinear module, leaving the learned SCN linear module (ablation) | SCN · MuJoCo | performance relative to standalone Linear policy: 68% with ES; 65% with PPO | Hurt |
| Add the linear control module to the nonlinear module (ablation) | SCN · MuJoCo and Roboschool | episodic reward: about 9% of the performance ... with ES, and 8% with PPO | Helped |
| Train linear and nonlinear components together rather than combining separately trained policies (ablation) | SCN · tested environments | performance relative to SCN: 18% ... with ES, and 21% with PPO | Helped |
| Replace the nonlinear MLP module with a CPG-inspired Fourier-series controller (ablation) | SCN locomotion variant · MuJoCo and Roboschool locomotion tasks | sampling efficiency and final reward | Helped |
| Retain the linear module alongside the CPG nonlinear controller (ablation) | Locomotor Net · locomotion tasks | performance | Helped |
| Use SCN-8 instead of MLP-8 (ablation) | Atari policy · Atari games | learning performance | Helped |
| Inject action and observation noise at test time | SCN compared with MLP-64 · test-time robustness | episodic reward | Helped |
Closest papers by meaning
- NerveNet: Learning Structured Policy with Graph Neural Networks (ICLR 2018 · similarity 0.91)
- Efficient Reinforcement Learning by Discovering Neural Pathways (NeurIPS 2024 · similarity 0.87)
- LipsNet++: Unifying Filter and Controller into a Policy Network (ICML 2025 · similarity 0.86)
- Analyzing Generalization in Policy Networks: A Case Study with the Double-Integrator System (AAAI 2024 · similarity 0.86)
- The Ladder in Chaos: Improving Policy Learning by Harnessing the Parameter Evolving Path in A Low-dimensional Space (NeurIPS 2024 · similarity 0.85)
- Adaptive Surrogate Gradients for Sequential Reinforcement Learning in Spiking Neural Networks (NeurIPS 2025 · similarity 0.85)
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.