Evidence item

The official SAIL page reports average simulated success rising from 25% to 73% across six tasks at a 45-candidate budget. DIRECT_TEXT

Authenticity UNVERIFIED · extraction UNVERIFIED · independence group none (unreviewed) · created seq 23784

Source: SAIL: Test-Time Scaling through Iterative Refinement for VLMs as Robot Trajectory Generators · locator {} · excerpt hash sha256:ef708acb704a89195235b32b7f797955e99776fa181070bb526dae4c593c79c9

We evaluated six manipulation tasks in the ALOHA simulator, with 20 initial configurations per task. Increasing the search budget raised the average success rate from 25% with a single generated trajectory to 73% with a budget of 45 candidate trajectories (up to 45 search nodes).