Run robot inference without running the GPUs.

$pip install dropbear

with dropbear.connect(model="molmoact2-droid") as policy:
    policy.run(
        observe=robot.get_observation,
        act=robot.execute_action,
        max_actions=250,
    )

Cloud inference, on the robot.

MolmoAct2 controlling an SO-101 smoothly through Dropbear cloud inference.

Download the SO-101 demo video

Models on Dropbear

A growing catalog of vision-language-action and world-action models for simulation and real robots.

ABot-M0-LIBERO
Cosmos3-Edge-Policy-DROID
Cosmos3-Nano-Policy-DROID
DreamZero-DROID
DreamZero-YAM
Fast-WAM
G0-VLA
Galaxea-G0.5-LIBERO
Galaxea-G0.5-SO101
Giga-World-Policy-0.5
GR00T-N1.7-3B
GR00T-N1.7-3B-LIBERO
Hy-Embodied-0.5-VLA-RoboTwin
LingBot-VA
LingBot-VLA
MolmoAct2-BimanualYAM
MolmoAct2-DROID
MolmoAct2-LIBERO
MolmoAct2-SO101
MolmoAct2-Think
MolmoAct2-Think-LIBERO
MolmoBot-DROID
pi0
pi0-FAST
pi0-FAST-LIBERO
pi0-LIBERO
pi0.5
pi0.5-LIBERO
RLDX-1-FT-LIBERO
RLDX-1-PT
Spirit-v1.5
VLA-JEPA
VLA-JEPA-LIBERO
WALL-OSS-0.5
Xiaomi-Robotics-0-LIBERO

Join the Dropbear beta

Tell us what you are building and which model you want to run.

What do you do?

Why are you interested in Dropbear?

Model you want to run