LOGIV preprint reports improved robot task completion through dependency checks

Researchers report that LOGIV, a framework that checks whether physical conditions needed for later robot actions remain valid, increased successful trials from 12/30 to 17/30 across three real-robot tasks in a preprint. The framework connects planning, observation and action through an explicit task graph. After a failure, it traces dependencies to identify what needs restoring and validates the revised continuation before execution resumes. It works above existing robot policies without adding trainable components. Across 20 dual-arm tasks from RoboTwin V2 and BiCoord, mean success without injected disturbances improved by 10.67 and 15.33 percentage points, respectively. Real-robot testing covered only three tasks.