RAMEN G1 IKEA Assembly — Models
Team RAMEN's RAMEN-Ori policies and YOLO-OBB detector for the IROS 2026 Humanoid IKEA Assembly Challenge (Unitree G1 + Dex1-1).
Robotics • UpdatedNote RAMEN-Ori policy trained on all six skills (checkpoints every 10k up to 600k). The starting point of the per-skill models.
Team-RAMEN/ramen-g1-ikea-assembly-ramen-ori-rotate-table-base
Robotics • UpdatedNote rotate_table_base, fine-tuned for the competition's slow arm. Our default at the competition (2026-09-30).
Team-RAMEN/ramen-g1-ikea-assembly-ramen-ori-pick-table-leg
Robotics • UpdatedNote pick_table_leg, fine-tuned for the competition's slow arm. Our default at the competition (2026-09-30).
Team-RAMEN/ramen-g1-ikea-assembly-ramen-ori-insert-table-leg
Robotics • UpdatedNote insert_table_leg, fine-tuned for the competition's slow arm. Our default at the competition (2026-09-30).
Team-RAMEN/ramen-g1-ikea-assembly-ramen-ori-flip-table
Robotics • UpdatedNote flip_table, fine-tuned on demonstrations that follow one procedure. Not yet validated on the robot.
Team-RAMEN/ramen-g1-ikea-assembly-yolo-obb
Object Detection • Updated • 105Note YOLO-OBB detector for the head camera (7 classes). Draws the overlay that RAMEN-Ori takes as input.