VLABench tasks¶
A simulated Franka Panda at a table: the 36 VLABench tasks we keep, one frozen instance each, run in two modes. Every row plays the verifier's replay of our reference solution (VLABench's own expert, recorded on the frozen instance).
36Tasks
36Keep
0Drop
0Untriaged
36/36Tagged
36/36Demos local
Speed
No task matches these filters.
Capability coverage · About this benchmark · How to edit a task page