Dataset

Learning to investigate.

Successful scripted trajectories include the physical search, inspection and testing steps needed to gather task-relevant evidence.

3,211successful episodes

405.5hours of interaction

20 Hzcamera, state and action records

Each frame records three camera views, robot state, action and the current subtask. Episodes average 7.6 minutes, or about 9,100 control steps. Only successful episodes that pass recording checks are retained.

The scripted oracles can read the private specification. Their routes deliberately include information-gathering steps; these are demonstrations of investigative behavior, not evidence that the oracle itself resolves uncertainty.

Training and evaluation instances do not overlap. Evaluation holds out kitchen styles and selected task configurations, including stamp targets and bolt-chain configurations.

Demonstration dataset

Open table ↗

Demonstration dataset. Scripted information gathering in each oracle's route, which the oracle performs although it knows the hidden information. Episodes: successful demonstrations. Hours: interaction time at 20 Hz.
Task	Scripted information gathering	Episodes	Hours
Locked Storage	follows the token chain	278	33.2
Search Room	opens compartments nearest-first until the targets are seen	294	46.8
Blackout Search	carries the lamp along a nearest-first search	201	53.0
Painted Cubes	removes covers and turns each cube face by face	276	50.3
Marked Mugs	lifts and tilts each vessel to read its label	312	20.9
Unfamiliar Containers	some episodes make one or two wrong tries before opening	398	76.9
Puzzle Box	tries sliders, one to three of which are blocked	447	21.1
Stamp Composition	prints test impressions before the final one	363	32.6
Wobbly Stand	releases the ball, places a shim, and tests again	381	18.9
Odd Parcel	weighs parcels on the balance	261	51.7
Total		3,211	405.5

Credits

Who built RoboQuest.

RoboQuest is developed at DeCLaRe Lab, Nanyang Technological University.

Authors

Affiliation

DeCLaRe Lab
Nanyang Technological University

Built with

RoboCasa365 kitchens simulated in MuJoCo, with a Franka Panda arm on a mobile base.

Citation

Cite RoboQuest.

Until the paper is released, please cite the project page. Paper and dataset release links will be added when available.

BibTeX

@misc{roboquest2026,
  title        = {RoboQuest: A Benchmark for Goal-Directed Embodied Exploration},
  author       = {Liu, Renhang and Majumder, Navonil and Pala, Tej Deep and Poria, Soujanya},
  year         = {2026},
  howpublished = {\url{https://declare-lab.github.io/RoboQuest/}},
  note         = {DeCLaRe Lab, Nanyang Technological University}
}