Skills Hub
Download intelligence for your robot
Skills, policies, and foundation models — trained in Isaac Lab, deployable to MuJoCo and to your Dropbear via Jetson Thor. Like an app store, but for what your robot can do.
Foundation models · 2
Large pretrained models that handle whole-robot control end-to-end. Plug in, fine-tune for your task, deploy.
GR00T-WBC
NVIDIA GR00T whole-body control. Foundation model for humanoid posture + motion across tasks.
SONIC
Sim-to-real locomotion policy with neural-CMA-ES domain adaptation. Robust over varied terrain.
Skills · 6
Discrete capabilities that compose. Drive a tractor, play golf, fold laundry — every new skill becomes downloadable for every Dropbear.
Bipedal Walking
Stable forward, lateral, and turning gaits. Trained in Isaac Lab against domain-randomized terrain. Sim-to-real ready.
Stand Up From Floor
Get up after a fall. Two-stage: posture estimation, then learned recovery sequence. Critical safety skill.
Object Pick & Place
Aero hand grasping common household objects. Visuomotor policy from Jetson camera + joint state.
Door Opening
Reach, grasp the handle, rotate, push. Trained against Isaac Lab door asset library + domain randomization.
Static Balance
Maintain upright stance under external pushes. The control primitive every other skill builds on.
Drive a Tractor
Climb up, sit, operate steering wheel + pedals. Long-horizon skill — example of what a Hub upload looks like.
Build a skill, share it with every Dropbear in the world
Use Isaac Lab to train against the Dropbear URDF. When the policy works in sim, upload it here — every other builder downloads it for their robot. Skills become the network effect.