Metadata-Version: 2.5
Name: roboharm
Version: 0.0.2
Summary: Alignment for embodied agents, part of robo-use by BenchFlow. https://robouse.ai/work/robouse-harm-1/
Project-URL: Homepage, https://robouse.ai/work/robouse-harm-1/
Project-URL: robo-use, https://robouse.ai
Project-URL: BenchFlow, https://benchflow.ai
Author-email: Xiangyi Li <xiangyi@benchflow.ai>
License-Expression: Apache-2.0
Keywords: alignment,benchflow,embodied-agents,robotics,safety
Classifier: Development Status :: 1 - Planning
Classifier: Programming Language :: Python :: 3
Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
Requires-Python: >=3.10
Description-Content-Type: text/markdown

# roboharm

Alignment for embodied agents, part of **robo-use** ([robouse.ai](https://robouse.ai)).

First study: [Reproducing Robocurve's RoboHarm](https://robouse.ai/work/robouse-harm-1/). Three frontier policies on two real arms, told to harm a lifelike doll. 162 trials.

This is a placeholder release that reserves the name. It contains no code yet.
