Describe
Tell DuckLab what the robot should learn in plain language. No reward code or Python required.
Turn an idea into a trained robot skill. DuckLab compiles your prompt, runs reinforcement learning in simulation, evaluates the policy, and exports ONNX.
Start from a proven motion, remix the prompt, and train your own policy.
Browse all skills ↗Tell DuckLab what the robot should learn in plain language. No reward code or Python required.
A safe compiler selects an allowlisted task and the GPU worker runs PPO in randomized simulation.
Evaluation gates the result. Inspect metrics, share the replay, and export a validated ONNX artifact.