5/5 ๐งช Robot policy evaluation.
We deploy OSCAR for real-world robot policy evaluation using the episodes from RoboArena. Our extensive experiments show a high correlation โ between virtual assessment in our world model and real-world results, paving the way for a future where robot policies are evaluated entirely within generated environments.
1/5 ๐ Thrilled to open-source OSCAR ๐ค โ an action-conditioned world model for robotics. It generalizes across different robot embodiments with precise action controllability. All trained on a single GH200 GPU, and outperforms existing open-sourced baselines, which have larger model capacity and need more compute.
Everything is public:
๐ Paper: https://t.co/IQkeWVR2XL
๐ Project: https://t.co/rbJJiSQfPu
๐ป Code: https://t.co/cpp7zxNwqv
๐ค Robot data: https://t.co/fMUZ9Ic59v
๐ค Human data: https://t.co/ouFbu9KOJl
๐ค Weights: https://t.co/zIqTtySxtX
#Robotics #WorldModels #AI #OpenSource
4/5 ๐ Results.
With our cleaned data and kinematic skeleton conditioning, we finetune the Cosmos2.5-2B model on a single GH200 GPU โก. Our method delivers superior quality on action following, appearance, and motion consistency, pushing SOTA performance with a fraction of the compute. ๐