Search papers, labs, and topics across Lattice.
3
0
5
5
LLMs spontaneously exhibit collaborative behaviors like perspective-taking and theory of mind in embodied settings, suggesting a surprising capacity for modeling human collaborators without explicit training.
Training robots on perfect data is a recipe for disaster: COFFAIL offers a dataset of both successful *and* failed robot coffee-making attempts, revealing the messy reality of real-world manipulation.
LLMs can drive embodied robot cognition for structured tasks, but don't expect them to follow instructions or avoid hallucinating about success just yet.
