LetTokenBurn
Sign in
Verified86%Harbor: pendingAgentic / Tool Use

RoboQuest

RoboQuest: Generalist Physical Agents that Search, Inspect and Test

A benchmark for goal-directed embodied exploration with 10 mobile manipulation tasks requiring agents to actively acquire task-relevant information through physical interaction.

PaperDatasetHomepage

How to run

ltb · roboquest
$ npm i -g lettokenburn-cli
$ ltb bench info roboquest

External harness. No Harbor adapter has been generated for this benchmark yet. `ltb bench info` prints the known links.

Runs happen on your machine with your own subscription; only the score summary is pushed. CLI docs

Scores

5 sourced entries
HarnessSourceConfidenceLink
GPT-6 Astra23.2N/AIndependentmediumOct 7, 2026arxiv.org
Claude Opus 5.513.8N/AIndependentmediumOct 7, 2026arxiv.org
GPT-6.1 Sol12.2N/AIndependentmediumOct 7, 2026arxiv.org
Claude Fable 5.111.4N/AIndependentmediumOct 7, 2026arxiv.org
Gemini 3.8 Flash2.0N/AIndependentmediumOct 7, 2026arxiv.org

Every number links to where it was published. Low-confidence rows are muted and need per-cell verification.

Reproduced by the community

All runs
Loading community runs…

About

RoboQuest comprises ten mobile manipulation tasks in kitchen environments, organized into search, inspection, and testing families. It includes a dataset of 5,000 demonstration episodes.

Scoring is based on episode success rate. The best frontier agent achieved only 23% success, indicating significant challenges in autonomous exploration and decision-making under uncertainty.