可微分仿真器能否带来更好的策略梯度？ICLR 2026 论文再审视估计器切换与方差控制
Does "Do Differentiable Simulators Give Better Policy Gradients?" Give Better Policy Gradients?
AI 导读
一篇入选 ICLR 2026 的 arXiv 论文探讨可微分仿真器是否真能带来更好的策略梯度。作者指出基于 REINFORCE 置信区间的间断检测方法需任务专属超参数调优且样本效率低，并提出仅用单一超参数、在非平滑区域切换估计器的轻量测试 DDCG，以及在可微分机器人控制任务中无需显式间断检测即可稳定方差的逐步逆方差实现 IVW-H。
正文 · 原文
Abstract:In policy gradient reinforcement learning, access to a differentiable model enables 1st-order gradient estimation that accelerates learning compared to relying solely on derivative-free 0th-order estimators. However, discontinuous dynamics cause bias and undermine the effectiveness of 1st-order estimators. Prior work addressed this bias by constructing a confidence interval around the REINFORCE 0th-order gradient estimator and using these bounds to detect discontinuities. However, the REINFORCE estimator is notoriously noisy, and we find that this method requires task-specific hyperparameter tuning and has low sample efficiency. This paper asks whether such bias is the primary obstacle and what minimal fixes suffice. First, we re-examine standard discontinuous settings from prior work and introduce DDCG, a lightweight test that switches estimators in nonsmooth regions; with a single hyperparameter, DDCG achieves robust performance and remains reliable with small samples. Second, on differentiable robotics control tasks, we present IVW-H, a per-step inverse-variance implementation that stabilizes variance without explicit discontinuity detection and yields strong results. Together, these findings indicate that while estimator switching improves robustness in controlled studies, careful variance control often dominates in practical deployments.
| Comments: | ICLR2026 |
| Subjects: | Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Robotics (cs.RO) |
| Cite as: | arXiv:2604.18161 [cs.LG] |
| (or arXiv:2604.18161v2 [cs.LG] for this version) | |
| https://doi.org/10.48550/arXiv.2604.18161 arXiv-issued DOI via DataCite |
|
| Journal reference: | The Fourteenth International Conference on Learning Representations. ICLR 2026 |
Submission history
From: Ku Onoda [view email]
[v1]
Mon, 20 Apr 2026 12:23:48 UTC (6,225 KB)
[v2]
Sat, 24 Oct 2026 00:48:22 UTC (6,213 KB)
来源：arXiv cs.AI · arxiv.org