Rohin Shah
TalkRL: The Reinforcement Learning Podcast
00:00
How to Evaluate the Robustness of Collaborative Agents
Celitz: One way of measuring how good your agents are collaborating with humans is, well, you just like have them play with humans and see how well that goes. But this paper proposed a maybe simpler and more reproducible test that you can run more often. The idea just write some unit tests for the robustness of your agents. Write some cases in which you think the correct action is unambiguously clear. And then just see whether your agent does, in fact do the right thing a on those inputs.
Transcript
Play full episode
