Mahal Systems BALROG
Fits teams studying long-horizon decision making with partial credit; borrow the per-game progress metrics that give signal even when a run never finishes.
What it asks the agent to do
- play a roguelike or grid-world game from text or image observations
- explore, gather and survive over long episodes
- follow instructions in a text-based game
What to borrow from it
grading designsandbox design
Caveats
Games only, no work tasks; the repo link was not on the paper page (code is pointed to via balrogai.com) and the license was not seen.
Details
| Workloads | games, planning |
|---|---|
| Interaction | autonomous long horizon |
| Environment | api sandbox |
| End state checked | No |
| Tools | game action interface |
| Grading | programmatic check |
| Size | Not published |
| Data | synthetic |
| Personal data | none |
| License | unknown |
| Harness | own repo |
| Status | active |