Research map / Optimization / Optimizers
Do Differentiable Simulators Give Better Policy Gradients?
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 |
|---|---|---|---|
| Use first-order rather than zeroth-order gradients | Pushing trajectory optimization with soft springs (k=10) · Pushing | optimization performance | Helped |
| Use zeroth-order rather than first-order gradients | Pushing trajectory optimization with stiff springs (k=1000) · Pushing | optimization performance | Helped |
| Use robust interpolation between first-order and zeroth-order gradients | Pushing trajectory optimization · Pushing | optimization performance | Helped |
| Use first-order rather than zeroth-order gradients | Friction trajectory optimization · Friction | optimization performance | Mixed |
| Use robust interpolation between first-order and zeroth-order gradients | Friction trajectory optimization · Friction | convergence speed | Helped |
| Use zeroth-order rather than first-order gradients | Tennis policy optimization · Tennis | policy performance across initial conditions | Helped |
| Use robust interpolation between first-order and zeroth-order gradients | Tennis policy optimization · Tennis | convergence speed | Helped |
| Use zeroth-order or interpolated gradients rather than first-order gradients | Ball with wall landscape optimization · Ball with wall | convergence to a minimum | Helped |
| Use zeroth-order or interpolated gradients rather than first-order gradients | Angular momentum transfer landscape optimization · Angular momentum transfer | convergence to a robust minimum | Helped |
| Use first-order gradients in a stiff or chaotic system | Policy-gradient estimation · Stiff dynamics and chaotic systems | gradient-estimator variance | Hurt |
| Choose the interpolation weight using empirical variance alone | Gradient-estimator interpolation · Nearly or strictly discontinuous landscapes | gradient-estimate accuracy | Hurt |
| Constrain interpolation to maintain accuracy while minimizing estimated variance | AoBG · Nearly or strictly discontinuous landscapes | gradient-estimate accuracy | Helped |
Closest papers by meaning
- Does “Do Differentiable Simulators Give Better Policy Gradients?” Give Better Policy Gradients? (ICLR 2026 · similarity 0.93)
- A Baseline for Any Order Gradient Estimation in Stochastic Computation Graphs (ICML 2019 · similarity 0.89)
- Implicit Bias of Policy Gradient in Linear Quadratic Control: Extrapolation to Unseen Initial States (ICML 2024 · similarity 0.87)
- Adaptive-Gradient Policy Optimization: Enhancing Policy Learning in Non-Smooth Differentiable Simulations (ICML 2024 · similarity 0.87)
- Overcoming the Sim-to-Real Gap: Leveraging Simulation to Learn to Explore for Real-World RL (NeurIPS 2024 · similarity 0.85)
- Adaptive Horizon Actor-Critic for Policy Learning in Contact-Rich Differentiable Simulation (ICML 2024 · 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.