← Leaderboard

Spatial Forcing

VLA
OpenHelix Robotics · Oct 14, 2025

Training method from HKUST(GZ), Westlake and others that aligns a VLA’s intermediate features with a 3D foundation model (VGGT), applied to OpenVLA-OFT on LIBERO, π0 on RoboTwin and real robots, and π0.5 for its RoboDojo entry.

Type
VLA
Parameters
—
Backbone
–
Weights
Open
License
MIT
Commercial use
Allowed
Released
Oct 14, 2025
Robots
AgileX bimanual PiPER, Franka Panda (sim), ARX X5 (sim)
SO-101
–
LeRobot
–
Inputs
–
VRAM
–

Benchmark results

2 of 13 benchmarks
#34 of 548.0%
LIBEROSaturatedSimulation
#22 of 7098.5%, self-reported by the authors, not a leaderboard entry
Not reported on 11 benchmarks
Wrong or missing number? Report an error