Research map / Reinforcement learning / Visual and representation RL
Is Bang-Bang Control All You Need? Solving Continuous Control with Bernoulli Policies
Tim Seyde, Igor Gilitschenski, Wilko Schwarting, Bartolomeo Stellato, Martin Riedmiller, Markus Wulfmeier, Daniela Rus
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 |
|---|---|---|---|
| Replace Gaussian policy head with Bernoulli bang-bang head | PPO, SAC, MPO, DreamerV2 · DeepMind Control Suite (8 tasks) | return | No clear effect |
| Use Bernoulli policy with SAC on Cartpole tasks | SAC · Cartpole | return | Hurt |
| Use Bernoulli policy with DreamerV2 on Finger Spin | DreamerV2 · Finger Spin | return | Hurt |
| Add quadratic action penalty to reward | Bernoulli policy · Cartpole Swingup | return | Hurt |
| Distill converged Gaussian teacher into Bernoulli student via behavioral cloning | Bernoulli student · DMC domains (excluding Humanoid) | return | Mixed |
| Use Bernoulli instead of Gaussian under perturbed environment parameters (mass, morphology, friction) | MPO · Real-World RL Cartpole/Walker/Quadruped | return | No clear effect |
| Use Bernoulli instead of Gaussian under sensor disturbances at transfer | MPO · Real-World RL disturbances | normalized score | No clear effect |
| Use Bang-Bang policy on sparse-reward tasks without action cost | Bang-Bang policy · Sparse Cartpole and Walker | convergence speed | Improved |
| Use Bang-Bang policy on sparse-reward tasks with action cost | Bang-Bang vs Gaussian and Bang-Off-Bang · Sparse tasks with action penalty | return | Improved |
| Use Bang-Bang policy on dense tasks | Bang-Bang vs Gaussian and Bang-Off-Bang · Cartpole Swingup, Walker Walk, Quadruped Run (dense) | return | No clear effect |
| Use Bang-Bang policy on pendulum swing-up without action cost | Bang-Bang vs Gaussian · Pendulum swing-up | rise time | Improved |
| Use Bang-Bang policy on pendulum swing-up with quadratic action cost | Bang-Bang vs Gaussian · Pendulum swing-up with action cost | return | Hurt |
Closest papers by meaning
- Overcoming Slow Decision Frequencies in Continuous Control: Model-Based Sequence Reinforcement Learning for Model-Free Control (ICLR 2025 · similarity 0.87)
- Actor-Free Continuous Control via Structurally Maximizable Q-Functions (NeurIPS 2025 · similarity 0.86)
- EvoControl: Multi-Frequency Bi-Level Control for High-Frequency Continuous Control (ICML 2025 · similarity 0.86)
- Learning from Suboptimal Data in Continuous Control via Auto-Regressive Soft Q-Network (ICML 2025 · similarity 0.86)
- Oracle-Efficient Reinforcement Learning for Max Value Ensembles (NeurIPS 2024 · similarity 0.85)
- Continuous Q-Score Matching: Diffusion Guided Reinforcement Learning for Continuous-Time Control (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.
