Robotic Manipulation on CALVIN
2.55Average LengthOpenVLA
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| OpenVLA2026.03 | 2.55 | — | — | |
| OctoBackbone=Octo2026.04 | 2.74 | — | — | |
| OpenVLABackbone=OpenVLA2026.04 | 3.02 | — | — | |
| OpenVLA + Rule VerifierBackbone=OpenVLA, Verifier Type=Rule2026.04 | 3.18 | — | — | |
| OpenVLA + ReflexionBackbone=OpenVLA, Verifier Type=Reflexion2026.04 | 3.19 | — | — | |
| HELM (Octo)Backbone=Octo2026.04 | 3.21 | — | — | |
| OpenVLA H=32Backbone=OpenVLA, Context Horizon (H)=322026.04 | 3.24 | — | — | |
| OpenVLA + Ensemble (×5)Backbone=OpenVLA, Ensemble Size=52026.04 | 3.29 | — | — | |
| OpenVLA + LoRA (50K)Backbone=OpenVLA, Fine-tuning/Adaptation Method=LoRA (50K)2026.04 | 3.31 | — | — | |
| OpenVLA + Oracle Mem.Backbone=OpenVLA, Memory Type=Oracle2026.04 | 3.41 | — | — | |
| HELM-FwdBackbone=OpenVLA, Recovery Strategy=forward recovery2026.04 | 3.44 | — | — | |
| HELM (OpenVLA)Backbone=OpenVLA2026.04 | 3.58 | — | — | |
| OmniStream2026.03 | 3.89 | — | — | |
| BC+SACLeaf=N/A2024.11 | — | 0.01 | — | |
| CARTLeaf=9982024.11 | — | 1.02 | — | |
| OpenVLA + AdapterSup.=10%, Backbone=OpenVLA, Adaptation Strategy=Supervised adaptation, Fine-tuning method=Adapter2026.06 | — | — | 1.61 | |
| OpenVLA + LoRASup.=10%, Backbone=OpenVLA, Adaptation Strategy=Supervised adaptation, Fine-tuning method=LoRA2026.06 | — | — | 1.78 | |
| OpenVLA + QLoRASup.=10%, Backbone=OpenVLA, Adaptation Strategy=Supervised adaptation, Fine-tuning method=QLoRA2026.06 | — | — | 1.7 | |
| OpenVLA + Selective LoRASup.=10%, Backbone=OpenVLA, Adaptation Strategy=Supervised adaptation, Fine-tuning method=Selective LoRA2026.06 | — | — | 1.86 | |
| OpenVLA Zero-shotSup.=0, Backbone=OpenVLA, Adaptation Strategy=Zero-shot2026.06 | — | — | 1.32 | |
| SemiVLA + AdapterSup.=10%, Backbone=OpenVLA, Adaptation Strategy=Semi-supervised adaptation, Fine-tuning method=Adapter2026.06 | — | — | 2.13 | |
| SemiVLA + LoRASup.=10%, Backbone=OpenVLA, Adaptation Strategy=Semi-supervised adaptation, Fine-tuning method=LoRA2026.06 | — | — | 2.42 | |
| SemiVLA + QLoRASup.=10%, Backbone=OpenVLA, Adaptation Strategy=Semi-supervised adaptation, Fine-tuning method=QLoRA2026.06 | — | — | 2.31 | |
| SemiVLA + Selective LoRASup.=10%, Backbone=OpenVLA, Adaptation Strategy=Semi-supervised adaptation, Fine-tuning method=Selective LoRA2026.06 | — | — | 2.58 | |
| SkillTreeLeaf=642024.11 | — | 1.9 | — | |
| SkillTree (DC + Distilling)Leaf=642024.11 | — | 3 | — | |
| SkillTree (Distilling)Leaf=642024.11 | — | 2.42 | — | |
| SPiRLLeaf=N/A2024.11 | — | 2.77 | — | |
| VQ-SPiRLLeaf=N/A2024.11 | — | 2.1 | — |