Xiaokang(Koe) Ye

@koe_ye40329

Ph.D. student @UCSanDiego. Leading @simworld_ai.

following
followers

RT-SAFE's reported results: 94.1% reach the goal; 0.7% finish without a safety event

A post introduces RT-SAFE as a way to evaluate embodied AI safety while the world keeps moving.

7 Sources, 14d ago, updated 14d ago

Code4Scene benchmarks agents building and editing 3D scenes in Unreal Engine

The benchmark’s creators say they tested 14 agent configurations on 95 public cases. Spatial composition was every agent’s weakest construction category.

5 Sources, 20d ago, updated 20d ago

Fable 5.1 reportedly beats Astra 2-1 in real-time virtual tag

EnactraAI says the matchup took place in a New York City simulation Astra created. The world kept moving while the models thought, leading to some awkward collisions with pedestrians and cars.

7 Sources, 32d ago, first seen 32d ago

BuildingBench tests coding agents on turning real-world images into 3D buildings

EnactraAI says GPT-6 Astra achieved the highest score in its comparison at about 80% lower cost than Fable 5.1.

5 Sources, 34d ago, first seen 34d ago

Featured in

RT-SAFE's reported results: 94.1% reach the goal; 0.7% finish without a safety event

Code4Scene benchmarks agents building and editing 3D scenes in Unreal Engine

Fable 5.1 reportedly beats Astra 2-1 in real-time virtual tag

BuildingBench tests coding agents on turning real-world images into 3D buildings

Featured in

RT-SAFE's reported results: 94.1% reach the goal; 0.7% finish without a safety event

Code4Scene benchmarks agents building and editing 3D scenes in Unreal Engine

Fable 5.1 reportedly beats Astra 2-1 in real-time virtual tag

BuildingBench tests coding agents on turning real-world images into 3D buildings