Operating environments: Physical GR-1 tabletop research bench
Control modes: Autonomous
Container-to-container transfer - Research demonstration - 2025-03-27 - Transfer household objects between trays, plates, baskets, bowls and pans. - environment: Physical GR-1 tabletop research bench - Hardware: Fourier GR-1 upper-body manipulation configuration; Fourier dexterous hands; Head-mounted egocentric camera; model unspecified - Software: GR00T-N1-2B policy, task post-training unless specified; NVIDIA L40 inference; system2 approximately10Hz, actiongeneration120Hz - Control: autonomous - Conditions: No base locomotion established; 20Hz VIVE/Xsens human demonstration capture is training data, not autonomous inference rate - source: arxiv.org
Fill and close storage compartments - Research demonstration - 2025-03-27 - Place an object in a wooden chest, cabinet or drawer, then close the compartment. - environment: Physical GR-1 tabletop research bench - Hardware: Fourier GR-1 upper-body manipulation configuration; Fourier dexterous hands; Head-mounted egocentric camera; model unspecified - Software: GR00T-N1-2B policy, task post-training unless specified; NVIDIA L40 inference; system2 approximately10Hz, actiongeneration120Hz - Control: autonomous - Conditions: No base locomotion established; 20Hz VIVE/Xsens human demonstration capture is training data, not autonomous inference rate - source: arxiv.org
Machinery-part packing - Research demonstration - 2025-03-27 - Pick machinery parts and tools and put them in a yellow bin. - environment: Physical GR-1 tabletop research bench - Hardware: Fourier GR-1 upper-body manipulation configuration; Fourier dexterous hands; Head-mounted egocentric camera; model unspecified - Software: GR00T-N1-2B policy, task post-training unless specified; NVIDIA L40 inference; system2 approximately10Hz, actiongeneration120Hz - Control: autonomous - Conditions: No base locomotion established; 20Hz VIVE/Xsens human demonstration capture is training data, not autonomous inference rate - source: arxiv.org
Pour loose industrial components - Research demonstration - 2025-03-27 - Grasp a mesh cup and pour its screws and bolts into a plastic bin. - environment: Physical GR-1 tabletop research bench - Hardware: Fourier GR-1 upper-body manipulation configuration; Fourier dexterous hands; Head-mounted egocentric camera; model unspecified - Software: GR00T-N1-2B policy, task post-training unless specified; NVIDIA L40 inference; system2 approximately10Hz, actiongeneration120Hz - Control: autonomous - Conditions: No base locomotion established; 20Hz VIVE/Xsens human demonstration capture is training data, not autonomous inference rate - source: arxiv.org
Cross-hand cylinder transfer - Research demonstration - 2025-03-27 - Pick up a cylinder, pass it between hands and place it in a bin. - environment: Physical GR-1 tabletop research bench - Hardware: Fourier GR-1 upper-body manipulation configuration; Fourier dexterous hands; Head-mounted egocentric camera; model unspecified - Software: GR00T-N1-2B policy, task post-training unless specified; NVIDIA L40 inference; system2 approximately10Hz, actiongeneration120Hz - Control: autonomous - Conditions: No base locomotion established; 20Hz VIVE/Xsens human demonstration capture is training data, not autonomous inference rate - source: arxiv.org
Pretrained-policy shelf placement - Research demonstration - 2025-03-27 - Use both hands to transfer an object across the body and place it on a shelf. - environment: Physical GR-1 tabletop research bench - Hardware: Fourier GR-1 upper-body manipulation configuration; Fourier dexterous hands; Head-mounted egocentric camera; model unspecified - Software: Pretrained GR00T-N1-2B without task post-training; NVIDIA L40 inference - Control: autonomous - Conditions: No base locomotion established; 20Hz VIVE/Xsens human demonstration capture is training data, not autonomous inference rate - source: arxiv.org
Two-robot component transfer - Research demonstration - 2025-03-27 - Hand a mesh cup containing a cylinder to another robot, then separate the cylinder and remaining contents into bins. - environment: Physical GR-1 tabletop research bench - Hardware: Fourier GR-1 upper-body manipulation configuration; Fourier dexterous hands; Head-mounted egocentric camera; model unspecified - Software: GR00T-N1-2B policy, task post-training unless specified; NVIDIA L40 inference; system2 approximately10Hz, actiongeneration120Hz - Control: autonomous - Conditions: No base locomotion established; 20Hz VIVE/Xsens human demonstration capture is training data, not autonomous inference rate - source: arxiv.org