How do I…?

Short answers, grouped by stage of the pipeline. Each links to the full guide.

Get going

Try it without a GPU	Live demo · Demo & Colab
Train my first policy	Quickstart
Do everything in a browser — also on a remote GPU server	WBC Studio

Motion data

Train on LAFAN / GMR / my own clips	Prepare motion data
Spot bad clips before training	Prepare motion data (Check the clips) — Studio Data → Clip
Double the library with mirrored clips	Prepare motion data (--mirror)
Look up the NPZ arrays or CSV columns	Motion data formats

Train

Pick a task	Choose a task
Tune rewards, observations, sampling or PPO without code	WBC Studio (Preset)
Reuse a set of changes, e.g. state estimation	WBC Studio (modules)
Resume, extend or pin a checkpoint	Train a policy
Train on several GPUs	Train a policy (Multi-GPU)
Train on rough terrain	Wbc-G1-Terrain-Scan — rough terrain

Evaluate and deploy

Watch a policy on many clips at once	WBC Studio (Play)
Debug the reference and adaptive sampling in viser	Evaluate and visualize
Export ONNX + config.yaml for the robot	Export and deploy
Look up observation sizes	Observations · the run’s params/config.yaml

Extend

Add a robot from its MJCF / URDF	Add a robot from model files
Ship a robot as a package	Add a robot
Add a paper’s setup as a task	Add a task or preset
Script the Studio	Studio HTTP API

Something broke

Troubleshooting covers the common errors — missing tasks, empty motion library, DOF mismatch, GPU / GL errors, resume, Studio access and robot checks.