Repository navigation
Publish learned skill capsules and scene-specific installation - #12
Merged
Merged
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Question
Advance the physical grounding research from task metadata plus an authored adapter to an installable learned movement artifact.
Evidence
This release distills the teacher's joint-target computation into a 34,171-byte canonical NumPy RBF model. Task phases, waypoints, grasp/contact checks and recovery remain authored. A new frozen eight-scene evaluation produces 7/8 for both teacher and student, without teacher fallback; total episode wall time increases from 6.273 to 6.360 seconds. Historical benchmark inputs, physics, success rules and published evidence remain unchanged.
A 39,845-byte data-only capsule binds task source, grounding, model and ten runtime source files. The local studio imports it, runs native placement and identical arm-command replay with open jaws, and enables execution only for that exact qualified scene. Browser verification includes file import, scene-change invalidation, requalification and execution without retraining. The bundled button preserves exact JSON bytes; rejected replacement imports clear readiness.
One known default scene plus repeat passed native macOS network denial after a direct libc connection was rejected with EPERM. Extracting the final archive reproduces exact actions, sampled states, metrics and native success checks with preinstalled dependencies. These are known-scene integration checks, not unseen/general robot competence or calibrated physics.
Reproduction
The new report and source/evidence live under
topics/physical-grounding/runs/2026-10-07-skill-capsule-v1/. Download/extractrelease/simlab-v0.4.zip, installrequirements.lock.txtonce, then runpython -m simlab.studio --offlineand drop the bundled capsule.CAPSULES.mdcovers author/train/build/install/qualify/run.native-offline/check.pyreproduces the native deny-network check on macOS.The release includes the dataset, protocol, collector/experiment source snapshots, frozen evaluation, profiling, native controls, browser replay, manifest and final-archive reproduction receipt. No GPU, hardware execution, learned perception or human-video motor supervision is claimed.
Checks