LADDER
LADDER: A Long-Horizon Robot Manipulation Benchmark
A benchmark for long-horizon robot manipulation with 145 tasks scaling from 1 to 10 subgoals across three simulation suites and two robot embodiments.
How to run
$ npm i -g lettokenburn-cli$ ltb bench info ladder
External harness. No Harbor adapter has been generated for this benchmark yet. `ltb bench info` prints the known links.
Runs happen on your machine with your own subscription; only the score summary is pushed. CLI docs
Scores
5 sourced entriesEvery number links to where it was published. Low-confidence rows are muted and need per-cell verification.
Reproduced by the community
All runsAbout
LADDER evaluates zero-shot skill composition in robotics using 145 tasks across SimplerEnv-Long, RobotArena-Long, and RoboLab-Long. It scales task horizon from 1 to 10 subgoals while keeping environments fixed.
Scoring uses programmatic predicates for subgoal progress and full task completion. Low-level policies collapse on long horizons (~5% success), while orchestration improves performance to ~25-62% depending on the method.