NottyDuck policies
Curated ONNX motor-policy releases for NottyDuck, Jalil's desk-based social media coach built on the open-source MicroDuck platform.
Raw training checkpoints live in private per-run repositories. A policy is
promoted here only after simulator rehearsal and a documented evaluation. Files
use policies/<gesture>/<version>/policy.onnx.
The 61-dimensional observation contract and 14-action layout must remain
compatible with the Pollen Robotics MicroDuck runtime. Export through the
repository's scripts/export.py; hand-converted checkpoints may omit the
observation normalizer and are not deployment-ready.
No policy artifact is published in the initial scaffold. Start with the
desk-peck smoke run documented in the source repository.
Collection including reachjalil/nottyduck-policies
Collection
NottyDuck's public persona, trained motor policies, and transparent evaluation ledger. • 3 items • Updated