Event

SHRIMP: Iterative Correction of Robot Plans in Natural Language

1 reports · 1 sources

Mya Schroder and co-authors propose the SHRIMP system, which lets users automatically generate hierarchical robot primitive plans in natural language and iteratively revise them through re-prompting and explicit corrections. After each revision, users can first validate the plan in simulation and, once satisfied, execute it on the physical robot. The authors validated it with a user study involving 35 participants on tabletop kitchen task planning, and report that SHRIMP improved users' perceived sense of control and robot transparency; this is the authors' conclusion based on that study.