𝚐𝔪𝟾𝚡𝚡𝟾

NOD-TAMP: Generalizable Long-Horizon Planning with Neural Object Descriptors

“intelligent replay” ✨ 

NOD-TAMP is a TAMP-based framework that learns short manipulation trajectories from a few human demonstrations, adapts them with NOD features, and composes them to handle complex, long-duration tasks. outperforms prior NOD methods on new tabletop manipulation challenges requiring broad generalization and has been successfully applied to real-world tasks such as tool-use and precise insertions.

NOD-TAMP: Generalizable Long-Horizon Planning with Neural Object Descriptors

“intelligent replay” ✨ 

NOD-TAMP is a TAMP-based framework that learns short manipulation trajectories from a few human demonstrations, adapts them with NOD features, and composes them to handle complex, long-duration tasks. outperforms prior NOD methods on new tabletop manipulation challenges requiring broad generalization and has been successfully applied to real-world tasks such as tool-use and precise insertions.

https://nodtamp.github.io