# Structured Control Nets for Deep Reinforcement Learning

arXiv: https://arxiv.org/abs/1802.08311 · HTML: https://mlsanity.com/paper/1802.08311/

## Key facts

- Cited by 8 papers in the map (data through 14 Oct 2026).
- First posted on arXiv 22 Feb 2018.

## Built on by (most cited first, 8 of 8)

- [Recurrent Model-Free RL Can Be a Strong Baseline for Many POMDPs](https://mlsanity.com/paper/2110.05038/) (2021)
- [Bayesian Controller Fusion: Leveraging Control Priors in Deep Reinforcement Learning for Robotics](https://mlsanity.com/paper/2107.09822/) (2021)
- [DeepCPG Policies for Robot Locomotion](https://mlsanity.com/paper/2302.13191/) (2023)
- [MPC-Inspired Reinforcement Learning for Verifiable Model-Free Control](https://mlsanity.com/paper/2312.05332/) (2023)
- [A Bibliometric Analysis and Review on Reinforcement Learning for Transportation Applications](https://mlsanity.com/paper/2210.14524/) (2022)
- [A Quadratic Actor Network for Model-Free Reinforcement Learning](https://mlsanity.com/paper/2103.06617/) (2021)
- [Fully-fused Multi-Layer Perceptrons on Intel Data Center GPUs](https://mlsanity.com/paper/2403.17607/) (2024)
- [A Policy Gradient-Based Sequence-to-Sequence Method for Time Series Prediction](https://mlsanity.com/paper/2406.09643/) (2024)

## Abstract (arXiv)

In recent years, Deep Reinforcement Learning has made impressive advances in solving several important benchmark problems for sequential decision making. Many control applications use a generic multilayer perceptron (MLP) for non-vision parts of the policy network. In this work, we propose a new neural network architecture for the policy network representation that is simple yet effective. The proposed Structured Control Net (SCN) splits the generic MLP into two separate sub-modules: a nonlinear control module and a linear control module. Intuitively, the nonlinear control is for forward-looking and global control, while the linear control stabilizes the local dynamics around the residual of global control. We hypothesize that this will bring together the benefits of both linear and nonlinear policies: improve training sample efficiency, final episodic reward, and generalization of learned policy, while requiring a smaller network and being generally applicable to different training methods. We validated our hypothesis with competitive results on simulations from OpenAI MuJoCo, Roboschool, Atari, and a custom 2D urban driving environment, with various ablation and generalization tests, trained with multiple black-box and policy gradient training methods. The proposed architecture has the potential to improve upon broader control tasks by incorporating problem specific priors into the architecture. As a case study, we demonstrate much improved performance for locomotion tasks by emulating the biological central pattern generators (CPGs) as the nonlinear part of the architecture.