Low-level policy sampling on 5 typical action models (Pick, Place, Open, Close, Toggle)
62Success RateCABTO
Evaluation Results
| Method | Links | |
|---|---|---|
| CABTOPolicy Model=Molmo+cuRobo+APIs, Execution Contexts=with Contexts, VLM Backbone=GPT-4o, Maximum sample attempts=32026.03 | 62 | |
| Molmo+cuRoboPolicy Category=Hierarchical2026.03 | 46 | |
| CABTOPolicy Model=Molmo+cuRobo+APIs, Execution Contexts=w/o Contexts, VLM Backbone=GPT-4o, Maximum sample attempts=32026.03 | 44 | |
| ReKepPolicy Category=Hierarchical2026.03 | 42 | |
| APIsPolicy Category=Rule-based2026.03 | 40 | |
| OpenVLAPolicy Category=End-to-end2026.03 | 28 | |
| VoxPoserPolicy Category=End-to-end2026.03 | 22 |