Physical latent probing on OGBench Cube benchmark leaderboard · SOTA2 Research

Physical latent probing on OGBench Cube

0.2Joint Position MSEDINO-WM

Evaluation Results

Method	Links
DINO-WM2026.03		0.2	0.87	0.263	0.852	0.004	0.999	0.167	0.917	0.038	0.982	0.007	0.998	0.769	0.411	0.916	0.304	0.29	0.799
LeWM2026.03		0.33	0.742	0.818	0.386	0.003	0.998	0.952	0.213	0.048	0.976	0.002	0.999	0.963	0.224	0.973	0.164	0.525	0.584
PLDM2026.03		0.34	0.728	0.661	0.536	0.013	0.993	0.985	0.117	0.066	0.967	0.003	0.999	0.989	0.218	0.989	0.106	0.503	0.6
LeWM2026.03		0.352	0.706	1.021	0.095	0.018	0.991	0.98	0.124	0.121	0.938	0.007	0.997	1.019	0.087	0.996	0.062	0.592	0.477
PLDM2026.03		0.372	0.695	1.016	0.115	0.052	0.974	0.996	0.056	0.234	0.876	0.031	0.985	1.021	0.066	0.996	0.061	0.611	0.464
DINO-WM2026.03		0.96	0.808	0.792	0.763	0.024	0.996	3.317	0.828	0.114	0.943	0.085	0.991	1.596	0.257	4.223	0.176	1.162	0.725