Zero-shot Generalization on CALVIN
Avg. sequence length
- 2023-12-20 — GR-1: Avg. sequence length 3.06
- 2024-02-18 — 3D Diffuser Actor: Avg. sequence length 3.27
- 2024-06-27 — RoboUniView: Avg. sequence length 3.647
- 2024-08-26 — GR-MG: Avg. sequence length 4.04
| Rank | Model | Avg. sequence length | Extra Training Data | Paper | Code | Year |
|---|---|---|---|---|---|---|
| 1 | GR-MG | 4.04 | ✓ | GR-MG: Leveraging Partially Annotated Data via Multi-Modal Goal-Conditioned Policy | bytedance/GR-MG | 2024 |
| 2 | MoDE | 4.01 | ✓ | Efficient Diffusion Transformer Policies with Mixture of Expert Denoisers for Multitask Learning | intuitive-robots/MoDE_Diffusion_Policy | 2024 |
| 3 | RoboUniView | 3.647 | RoboUniView: Visual-Language Model with Unified View Representation for Robotic Manipulation | liufanfanlff/robouniview | 2024 | |
| 4 | 3D Diffuser Actor | 3.27 | 3D Diffuser Actor: Policy Diffusion with 3D Scene Representations | nickgkan/3d_diffuser_actor | 2024 | |
| 5 | GR-1 | 3.06 | ✓ | Unleashing Large-Scale Video Generative Pre-training for Visual Robot Manipulation | bytedance/gr-1 · bytedance/GR-MG · GR1-Manipulation/GR-1 | 2023 |