{"type":"rich","version":"1.0","provider_name":"Transistor","provider_url":"https://transistor.fm","author_name":"Eye on AI Weekly Research Watch","title":"WorldSample: Closed-loop Real-robot RL with World Modelling","html":"","width":"100%","height":180,"duration":224,"description":"Training robots via reinforcement learning is expensive because each physical trial is costly and only reveals one outcome path. WorldSample addresses this by combining real robot rollouts with a learned \"world model\" that generates additional high-fidelity synthetic experience, reducing the need for physical trials. A technique called Policy-Paced Learning carefully schedules how synthetic data is used to avoid compounding hallucination errors or overestimating value. On contact-rich manipulation tasks, WorldSample boosts success rates by 28% while cutting training steps by 59%. This approach could substantially lower the cost and time required to deploy capable RL-trained robots in real-world settings.\n\nAuthors: Yuquan Xue, Le Xu, Zeyi Liu, Zhenyu Wu, Zhengyi Gu, Xinyang Song, Bofang Jia, Ziwei Wang\n\nPaper: https://arxiv.org/abs/2607.02431v1","thumbnail_url":"https://img.transistorcdn.com/NzT0CgVgc5EYmTYVshPdpb6IAFKCteYvSiwlDGdBSuw/rs:fill:0:0:1/w:400/h:400/q:60/mb:500000/aHR0cHM6Ly9pbWct/dXBsb2FkLXByb2R1/Y3Rpb24udHJhbnNp/c3Rvci5mbS80ZDk4/YjBiMGUyYzJiNzIw/YTRjYjc4OTM2YzM4/OGQ5Ny5qcGc.webp","thumbnail_width":300,"thumbnail_height":300}