RAX: Open-Source Framework for Autonomous Robot Exploration and Skill Acquisition
清华姚班、哈佛学霸，让机器人自己写代码、练技能，终于不用人手把手带了
Researchers from Harvard, NVIDIA, UC Berkeley, MIT, and Stanford introduce RAX, an open-source framework that allows robot agents to autonomously explore and acquire skills through online experience. RAX enables robots to write and execute code in real or simulated environments, improving generalization and task performance across various robotic platforms. The framework shows significant improvements in benchmark tests, but real-world deployment and skill reuse still require further validation.
The paper introduces RAX, an open-source framework for robot agents that enables autonomous exploration and skill acquisition without human intervention. It demonstrates improved performance across multiple robotic benchmarks, but real-world deployment and skill generalization remain to be validated.
Source: Chinese robotics — Research news discovery · Read original article ↗
English translation unavailable. Choose the original article to read it.
What the source reports
Publisher-reported claims, with original evidence. These results have not been independently verified by RoboSignal.
Reported numbers
analysis scripts written
23
View original evidence
▲《Scaling Online Exploration for Robot Agents》 比如，在机器人执行开空气炸锅任务时，智能体一开始要边看深度相机，边写分析脚本，边试动作，花了44分钟，写了23个分析脚本才完成第一次抓取。
Open source S6time spent on task
44 minutes
View original evidence
▲《Scaling Online Exploration for Robot Agents》 比如，在机器人执行开空气炸锅任务时，智能体一开始要边看深度相机，边写分析脚本，边试动作，花了44分钟，写了23个分析脚本才完成第一次抓取。
Open source S6
- paper: not_reported
- code: not_reported
- weights: not_reported
- dataset: not_reported
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
▲《Scaling Online Exploration for Robot Agents》 比如，在机器人执行开空气炸锅任务时，智能体一开始要边看深度相机，边写分析脚本，边试动作，花了44分钟，写了23个分析脚本才完成第一次抓取。
Open source S6
在仿真实验中，RAX的平均表现比表现最好的基线模型高出 37.4%。
Open source S10
在机器人开空气炸锅任务中，智能体打开空气炸锅抽屉，将一根香蕉放入炸篮，再关闭抽屉。
Open source S22
智能体首先将RGB-D相机采集的观测数据转换为点云，并统一到机器人基座坐标系中。
Open source S23
然后，它利用自身测量得到的数据，构建空气炸锅的几何模型。
Open source S24
研究团队提出一个新的思路， 让智能体直接为机器人编写程序， 这样，整个过程不需要人工设计 动作原语 （ action primitives）。
Open source S16
智能体会将成功的交互过程，转化为可自主执行的代码程序，并保存可供后续实验和任务复用的知识。
Open source S20
Source：Chinese robotics — Research news discovery · news.qq.com