Assistax
收藏资源简介:
Assistax是一个开源的硬件加速强化学习基准测试,专为辅助机器人任务而设计。它使用JAX的硬件加速功能,在物理仿真中实现了显著的学习速度提升。Assistax采用多智能体强化学习来模拟辅助机器人与活跃人类患者的交互,并测试机器人在零样本协调能力方面的表现。该数据集支持单智能体强化学习(SARL)、多智能体强化学习(MARL)和零样本协调(ZSC)算法,在一个具有连续动作和硬件加速的3D环境中进行。Assistax的创建过程包括环境设计、算法实现和优化计算,旨在提高辅助机器人强化学习研究的效率。该数据集的应用领域主要是辅助机器人,旨在解决机器人与人类交互中的复杂性问题,例如协调和合作。
Assistax is an open-source hardware-accelerated reinforcement learning benchmark specifically designed for assistive robotics tasks. It leverages the hardware acceleration capabilities of JAX to achieve significant learning speedups in physical simulations. Assistax employs multi-agent reinforcement learning to simulate the interactions between assistive robots and active human patients, and evaluates the robots' zero-shot coordination performance. This benchmark supports single-agent reinforcement learning (SARL), multi-agent reinforcement learning (MARL), and zero-shot coordination (ZSC) algorithms, and operates in a 3D environment with continuous action spaces and hardware acceleration. The development of Assistax encompasses environment design, algorithm implementation, and computational optimization, aiming to improve the efficiency of reinforcement learning research for assistive robotics. This benchmark is primarily targeted at assistive robotics applications, aiming to address the complexity challenges in robot-human interaction, including coordination and cooperation.
Assistax 数据集概述
📌 数据集简介
- 名称:Assistax
- 领域:辅助机器人学
- 内容:提供硬件加速环境及基线算法实现,用于强化学习(RL)和多智能体强化学习(MARL)训练
- 技术栈:基于JAX和Brax实现快速训练流程
📄 论文链接
🏗️ 环境配置
- 依赖管理工具:推荐使用
uv - 安装步骤: bash git clone https://github.com/assistive-autonomy/assistax.git cd assistax uv sync && uv pip install -e .
� 环境示例
- Scratch:机器人需在人类手臂随机目标位置施加指定力度
- Bed Bath:机器人需按顺序擦拭人类手臂上的所有目标点
- Arm Assist:机器人需协助人类将手臂抬回舒适位置
📊 基线算法
| 算法 | FF | PS | NPS | RNN |
|---|---|---|---|---|
| IPPO | ✅ | ✅ | ✅ | ✅ |
| MAPPO | ✅ | ✅ | ✅ | ✅ |
| ISAC | ✅ | ❌ | ✅ | ❌ |
| MASAC | ✅ | ❌ | ✅ | ❌ |
🧠 预训练策略
- 获取方式:从Hugging Face下载
zoo.tar.gz - 配置路径:修改
assistax/baselines/ZSC/config/ppo_aht.yaml中的ZOO_PATH
📚 相关资源
🖋️ 引用格式
bibtex @misc{hinckeldey2025assistaxhardwareacceleratedreinforcementlearning, title={Assistax: A Hardware-Accelerated Reinforcement Learning Benchmark for Assistive Robotics}, author={Leonard Hinckeldey and Elliot Fosong and Elle Miller and Rimvydas Rubavicius and Trevor McInroe and Patricia Wollstadt and Christiane B. Wiebel-Herboth and Subramanian Ramamoorthy and Stefano V. Albrecht}, year={2025}, eprint={2507.21638}, archivePrefix={arXiv}, primaryClass={cs.AI}, url={https://arxiv.org/abs/2507.21638}, }

- 1Assistax: A Hardware-Accelerated Reinforcement Learning Benchmark for Assistive Robotics爱丁堡大学, 本田欧洲研究院, DeepFlow · 2025年



