LiMA
“LiMA introduces a dual-system architecture that separates high-level, long-horizon planning (the 'Future Dreamer') from high-frequency, reactive motor control (the 'Motion Refiner').”
Source→“LiMA reduces inference latency by 45.8% compared with Cosmos-Policy via asynchronous decoupling”
Source→“The Dreamer is initialized from a pre-trained Cosmos-Predict2-2B [56] to inherit physical world priors”
Source→“The multimodal perception suite comprises three Intel RealSense D435 cameras: one head-mounted camera capturing the global ego-centric view and two wrist-mounted cameras capturing localized coordination views”
Source→“Our real-world evaluation platform consists of a bimanual setup featuring two 6-DoF UR5 robotic arms”
Source→“human finger configurations are recorded via MetaGlove Pro gloves and mapped onto the 22-DoF SharpaWave hands through a joint-space kinematic retargeting pipeline”
Source→“we use VIVE Trackers to capture the relative spatial trajectories of the human wrists for real-time end-effector control of the UR5 arms”
Source→AI-extracted from podcast / newsletter / paper summaries. May contain errors.