research RobotArena ∞ Unlimited robot benchmarking via real-to-sim translation — ICLR 2026 ParticleDiffuser Generative 3D object-particle dynamics for planning OSWorkerBench 100 long-horizon office workflows for computer-use agents, released with Tencent's UI-Mate SCAP Transductive test-time adaptation via supportive clique-based attribute prompting — CVPR 2025 D2ART Dual-modal prompting for vision–language models at test time engineering Multi-Skill Dexterous Manipulation From RL to flow-matching — a three-stage pipeline for dexterous grasping Throughline A goal-tracking planner that turns notes and meetings into todos