The paper presents Language Instruction grounding for Motion Planning (LIMP), a system that enables robots to follow complex language instructions by generating semantic maps. LIMP uses foundation models and temporal logics to create these maps, which help robots follow long-horizon instructions with open vocabulary referents and complex spatiotemporal constraints. The system was tested in three real-world environments, with successful results in 90% of object-goal navigation and 71% of mobile manipulation instructions.
Publication date: 20 Feb 2024
Project Page: robotlimp.github.io
Paper: https://arxiv.org/pdf/2402.11498
