Stable-FastTD3 Push Task Dataset: Stability and Reward Curves of Four Algorithms (Stable-FastTD3, MPC-only, RoHM-only, FastTD3)
收藏IEEE2026-04-17 收录
下载链接：
https://ieee-dataport.org/documents/stable-fasttd3-push-task-dataset-stability-and-reward-curves-four-algorithms-stable
下载链接
链接失效反馈官方服务：
资源简介：
This dataset corresponds to the push task experiments in the paper \Stable-FastTD3 Integrating Stability-Aware Fusion: A Multi-Task Evaluation on HumanoidBench\. It contains processed stability and reward curves for four reinforcement learning algorithms (Stable-FastTD3, MPC-only, RoHM-only, FastTD3) applied to humanoid robot control in the HumanoidBench push task. Data were collected via parallel environment sampling (128 environments) over 1 million steps, averaged across three random seeds, and processed with sliding window smoothing and custom transformations. It supports quantitative comparison of algorithm performance and stability analysis for contact-rich manipulation tasks.
提供机构：
Rongzhao Wang
