MMODELYST
Evals/BabyAI
EVL

BabyAI

Grounded instruction env

Source
RL environmentAgentsSuccess rate on instruction-following levels
What this measures

A gridworld where an agent follows simple language instructions ('go to the red ball, then open the green door'). A controlled testbed for grounded language understanding.

Sample
'Pick up the blue key and open the door on your left.' — solved with low-level move/turn/pickup actions.
Leaderboard · 0 models
Success rate on instruction-following levels — higher is better
No scores recorded yet.
What improves this score

Datasets & environments shown to raise it.

No data yet.
Research using this benchmark
all →

Matched by name in title/abstract.