Event
RGraph-Eval: Evaluating Robot Video Generation with Task Scene Graphs
1 reports · 1 sources
Aleksandr Lazarev and co-authors propose and release RGraph-Eval, a framework for evaluating robot video generation based on reference demonstrations. It uses task scene graphs to explicitly expose interaction errors in generated videos, such as those involving objects, grasping, and placement, and outputs interpretable scores that include visible objects, relationship changes, and 3D path shapes. The framework uses a foundation-model pipeline to identify manipulation roles, locate object masks, infer and verify state transitions and action intervals, and reconstruct the 3D motion of manipulated objects. The authors state that the code will be made public, but no specific address is given.
Report timeline
- RGraph-Eval: Evaluating Robot Video Generation with Task Scene Graphs
The authors propose RGraph-Eval, a reference-demonstration-based framework for evaluating robot videos that uses task scene graphs to explicitly expose interaction errors in generated videos, such as errors involving objects, grasps, and placements. The framework uses a foundation-model pipeline to identify manipulation roles, locate object masks, infer and verify state transitions and action intervals, and reconstruct the 3D motion of manipulated objects. It outputs an interpretable score that covers visible objects, relation changes, and 3D path shape.
arXiv cs.RO ·