Towards Zero-Knowledge Task Planning via a Language-based Approach
Liam Merz Hoffmeister, Brian Scassellati, Daniel Rakita · 2026
Read Towards Zero-Knowledge Task Planning via a Language-based Approach by Liam Merz Hoffmeister, Brian Scassellati, Daniel Rakita (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2601-03398.txt.
Description
Liam Merz Hoffmeister, Brian Scassellati, and Daniel Rakita formalize a hard version of robot planning: zero-knowledge task planning. In this problem, the system must build a sequence of actions for a goal, without any task-specific knowledge or setup. Their approach leans on a large language model. The model turns natural language instructions into subtasks, and then into behavior trees that a robot can execute.
When execution errors arise, the same model adjusts the behavior trees on the fly, in a refinement loop, rather than failing outright. Experiments in the AI2-THOR household simulator show the approach improves overall task performance. It beats alternatives that do use task-specific knowledge. The authors present the work as a first investigation. It shows language models can cover several parts of the zero-knowledge problem, from decomposition to repair.
Catalog record
| CREATOR | Liam Merz Hoffmeister, Brian Scassellati, Daniel Rakita |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 32 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2601-03398