List your company

Embodied Reasoning on EB-Habitat

73.33Base Success RateMuSix
52.5357.9363.3368.73Jul 1, 2026

Evaluation results

MethodLinks
MuSixbackbone=Qwen3-VL-4B-Instruct, world model groups (G)=3, world models (K)=15, top-k models=4, alpha_max=0.3, gamma=1.02026.0773.3326.6737.3342.673626.6740.44
SayCanPay2026.077211.6730.6737.3337.3317.3334.39
FLAREparadigm=retrieval-augmented generation2026.0772418.6729.332810.6727.11
Conventional MoErouting=standard top-k routing2026.076010.6783230.671225.56
LLM-Plannerparadigm=few-shot prompting without explicit world modeling2026.0753.332.6718.6722.6730.6710.6723.11