The residual library and self-review both improve success in simulation. Without execution feedback, the four variants achieve:
| Objective representation | Without self-review | With self-review |
|---|---|---|
| VLM-generated expressions | 8/40 (20%) | 14/40 (35%) |
| Residual library | 20/40 (50%) | 26/40 (65%) |
With the library, self-review, and at most two feedback rounds, cumulative simulation success reaches 36/40 (90%), allowing up to three executions per trial.
On the physical robot, selected simulation-successful programs succeed in 32/40 executions (80%) across two perturbed scene configurations. See the paper for the full protocol, baselines, and limitations.




