← Leaderboard

HoloBrain-0

VLA
Horizon Robotics · Feb 12, 2026

VLA that feeds embodiment priors (camera parameters and URDF kinematics) into the model for 3D spatial reasoning, in a 1.1B QW and a 0.2B GD variant, with the RoboOrchard training stack. It reports LIBERO and RoboTwin 2.0; nothing ties it to the existing Table30 “holobrain” entry.

Type
VLA
Parameters
1.1B
Backbone
Qwen2.5-VL-3B (truncated, QW variant); Grounding DINO Tiny (0.18B GD variant)
Weights
Open
License
Apache-2.0
Commercial use
Allowed
Released
Feb 12, 2026
Robots
Franka Panda (sim), Aloha-AgileX (sim)
SO-101
–
LeRobot
–
Inputs
–
VRAM
–

Benchmark results

2 of 13 benchmarks
RoboTwin 2.0Near saturationSimulation
  • 92.1%, self-reported by the authors, not a leaderboard entryPaper / model card ↗Feb 2026· self-reported, 50 clean + 500 randomized demos· n = 10000· non-standard
92.1%, self-reported by the authors, not a leaderboard entrynot ranked
LIBEROSaturatedSimulation
#34 of 7097.4%, self-reported by the authors, not a leaderboard entry
Not reported on 11 benchmarks
Wrong or missing number? Report an error