KnowNo — Robots That Ask for Help

A research framework that helps a robot ask for clarification when a request has multiple plausible interpretations, such as which bowl to put in a microwave. KnowNo evaluates uncertainty-aware planning in simulation and on physical robots. Evidence and limits: The paper’s guarantees assume correctly grounded scene descriptions, successful low-level execution and faithful human help. They are not overall safety guarantees covering perception errors, action failures or incorrect human answers. We did not run the project. Original project images and demonstrations (with attribution):

Sources: https://robot-help.github.io/ https://arxiv.org/abs/2307.01928 https://arxiv.org/html/2307.01928v2 https://github.com/google-research/google-research/tree/master/language_model_uncertainty https://youtu.be/xCXx09gfhx4