3D Office Agent
An agent that completes routine office tasks inside a 3D floor plan, and leaves a log of every action and observation so any run can be replayed.
How it works
What it does, and what it does not
- 01
Maps an instruction such as “bring this to the meeting room” onto a navigable 3D floor plan.
- 02
Plans a route, acts step by step and records each action and observation in a replayable log.
- 03
Reports task completion on a fixed set of seven routine tasks.
- Tasks
- 7 routine, fixed set
- Floor plan
- grid 8 × 6, 3 floors
- Log
- every action + observation
- Stage
- public review
Known limits
- Tested on one floor plan family; a new building needs its own map.
- Simulated environment only. It does not control physical devices.
Version and changelog
v0.6.0, and how it got here
Semantic versions. Changes that came out of review name the review that asked for them.
- v0.6.02026-10-09Current
- AddedReplay viewer for published task logs.
Review record
Internal → Community → Public
Every stage, its dates and what reviewers said, including the requests for changes. Each comment names the version it read.
Internal review (Institute)
2026-09-02 → 2026-09-08Passed to community- Editorial BoardInstitute
Runs from the repository; logs replay to the same final state.
✓ Accepted
Community review (Community)
2026-09-09 → 2026-10-072 accepted- T. AdityaReviewer · evaluation
Task completion is reported per task with failures listed. Accepting.
✓ Accepted
Public review (Public)
2026-10-08 → 2026-10-29Open for commentNo reviews filed in this stage yet.
Cite this
Cite the version you used
Results can change between versions. Cite the version and the proof ID so a reader can open exactly what you saw.
Nugroho, F., Tan, A. (2026). 3D Office Agent (v0.6.0) [Software]. Adaptive Institute. AI-PRF-0014. https://adaptive-institute.id/products/3d-office-agent