Overview
Leaderboard
Every model climbs the same ladder. Its height is the sum of its success rates on the four levels, so a perfect policy stands on top of rung 4 and every bit of progress on a harder level lifts it further.
What Makes a Rung Harder?
Each family of skills is instantiated at four levels. Same embodiment, same physics — only the demands grow.
Task Ladder
All tasks, laid out by level, physics domain and embodiment. Hover to play, click to open.
Results
Success falls off as models climb: the ladder separates what current policies can do from what remains open.
Citation
@article{maniladder2026,
title = {ManiLadder: Benchmarking Robot Manipulation Through a
Categorized and Multi-Difficulty Level Task Ladder},
author = {Coming soon},
year = {2026}
}