Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension


Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
3 changes: 2 additions & 1 deletion .github/workflows/validate.yml
Original file line number Diff line number Diff line change
Expand Up @@ -16,7 +16,8 @@ jobs:
- uses: actions/setup-node@49933ea5288caeca8642d1e84afbd3f7d6820020 # v4
with:
node-version: '22'
- run: npm ci --ignore-scripts
- run: npm test
- run: npm run validate
- run: npm run build
- run: git diff --exit-code -- docs/topics.json
- run: git diff --exit-code -- docs
1 change: 1 addition & 0 deletions .gitignore
Original file line number Diff line number Diff line change
Expand Up @@ -7,3 +7,4 @@ credentials*
outputs/
*.log
docs/topics.json.tmp
.wrangler/
4 changes: 2 additions & 2 deletions CONTRIBUTING.md
Original file line number Diff line number Diff line change
@@ -1,11 +1,11 @@
# Contribute one testable question

Start with a [topic](README.md#choose-a-question) and its numbered open questions. All current topics are **scoping**; every baseline is **not run**. No owner or result is implied by an open question.
Start with a [topic](https://github.com/Dvidia-Inference/dvidia-research#choose-a-question) and its numbered open questions. Topic metadata remains **scoping** and the original baseline protocols are **not run**. Separately recorded source reviews, a Clef feasibility pilot and a CPU contact fixture are linked from the journal. No owner or result is implied by an open question.

## Pick a small entry point

1. Open a [research question](https://github.com/Dvidia-Inference/dvidia-research/issues/new?template=research-question.yml) for a source correction, missing definition or testable uncertainty. Link the exact topic and question ID.
2. For a run, open an [experiment issue](https://github.com/Dvidia-Inference/dvidia-research/issues/new?template=experiment.yml) and copy the [protocol template](templates/experiment.md) into the topic's `experiments/` folder with the next available number.
2. For a run, open an [experiment issue](https://github.com/Dvidia-Inference/dvidia-research/issues/new?template=experiment.yml) and copy the [protocol template](https://github.com/Dvidia-Inference/dvidia-research/blob/main/templates/experiment.md) into the topic's `experiments/` folder with the next available number.
3. Use a [topic proposal](https://github.com/Dvidia-Inference/dvidia-research/issues/new?template=topic-proposal.yml) only when the question does not fit the four existing topics. Explain the smallest experiment that would resolve it.

Useful first work includes checking a citation, annotating a few permission-cleared examples, adding a malformed-file fixture, or documenting a failed reproduction. A large training run is not a prerequisite.
Expand Down
23 changes: 23 additions & 0 deletions DEPLOYMENT.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,23 @@
# Research site release

The static site has a dedicated Cloudflare Worker, `dvidia-research`, serving only `research.dvidia.org`. The configuration has no application bindings and does not modify the main DVIDIA Worker. Custom Domains manages the hostname's DNS and certificate.

Build and review before releasing:

```sh
npm ci --ignore-scripts
npm run check
npm test
node scripts/cloudflare-release.mjs dry-run
```

Merge reviewed source and generated `docs/` output to `main`, then deploy the same revision:

```sh
node scripts/cloudflare-release.mjs deploy
node scripts/cloudflare-release.mjs status
```

The release helper uses the existing local DVIDIA Cloudflare credential, verifies the account and exact hostname, and passes the token only to Wrangler's process environment. Do not add credentials to this repository. Wrangler 4.112.0 was used for this release; Node 22 runs the site generator. Research readers have no hosted API or external font dependency.

Verify HTTPS for the journal, a paper, its Markdown download and RSS. Check that a nonexistent path returns 404 and that a paper is readable without JavaScript. Record the deployed version ID separately from source evidence. Rollback uses Wrangler's version rollback for this dedicated Worker; review its current version history first. A rollback must not target the main `dvidia` Worker.
15 changes: 11 additions & 4 deletions README.md
Original file line number Diff line number Diff line change
Expand Up @@ -2,9 +2,9 @@

Small, reproducible studies of what makes a physical demonstration useful: visible evidence, defensible relationships, traceable data and measured physical grounding.

**Status: scoping. No experiments have run.** This repository contains source reviews, open questions and proposed protocols. It does not claim trained robot skills, validated transfer or measured improvements.
**Status: source reviews and bounded local experiments.** Five licensed POV excerpts have been processed in an image-API feasibility pilot, and 54 native CPU contact-fixture trials have been recorded. The original baseline protocols, GPU training and physical robot transfer remain unrun.

[Browse the research site](https://dvidia-inference.github.io/dvidia-research/) · [Read the research agenda](reports/Physical%20skill%20research%20agenda.md) · [Start contributing](CONTRIBUTING.md)
[Browse the research site](https://research.dvidia.org/) · [Read the research agenda](reports/Physical%20skill%20research%20agenda.md) · [Start contributing](CONTRIBUTING.md)

## Choose a question

Expand All @@ -20,6 +20,10 @@ All four covers are **AI-generated editorial illustrations**, not footage, measu

## Latest study

[Fast accurate robot training simulation](reports/Fast%20accurate%20robot%20training%20simulation.md) translates observation-to-skill research into an offline training-environment design. The published [contact fixture](topics/physical-grounding/runs/2026-10-07-contact-fixture/) contains byte-identical benchmark code and recorded outputs for 54 CPU trials. Contact-force fidelity, deformable calibration, GPU throughput and physical transfer remain separate gates.

[Grok robotics research on X](reports/Grok%20robotics%20research%20on%20X.md) checks the initial thesis against primary literature. [Spatially grounded downloadable robot skills](reports/Spatially%20grounded%20downloadable%20robot%20skills.md) sets out the proposed controller, compatibility and evaluation contract.

[Clef and grounded video observations](reports/Clef%20and%20grounded%20video%20observations.md): five licensed POV excerpts processed, with actual costs, timings and important caption/checker failures. This is an engineering feasibility result; the separate 40-episode accuracy pilot still needs its reference set.

[From Skillspace to robot skill](reports/From%20Skillspace%20to%20robot%20skill.md) reviews Argus and proposes a measurable path from accepted videos to robot-tested releases. Includes a current implementation gap analysis, DOF compatibility requirements and an unrun [annotation pilot](topics/evaluation-provenance/experiments/0002-skill-graduation.md).
Expand All @@ -34,15 +38,18 @@ The four [starter issues](https://github.com/Dvidia-Inference/dvidia-research/is

## Preview and maintain the site

Use Node.js 22 or newer; there are no package dependencies to install.
Use Node.js 22 or newer. The pinned build-time Markdown parser is installed by `npm ci`; readers require no JavaScript or external font service.

```sh
npm ci --ignore-scripts
npm run check
npm test
python3 -m http.server 8201 --directory docs
```

Open `http://localhost:8201`. Edit each topic's metadata and bibliography, then run `npm run build` to regenerate `docs/topics.json`. Commit both the source edits and generated catalog. GitHub Pages serves `docs/` from `main`; the validation workflow checks references, metadata, safe links and deterministic output.
Open `http://localhost:8201`. Edit report Markdown and `publications.json` for publications; edit topic metadata and bibliographies for questions. `npm run build` generates the topic catalog, six publication readers, four topic readers, portable downloads, RSS and sitemap. Commit source edits and generated output. The static `docs/` directory supports the custom research subdomain and a subfolder preview through relative links. Validation checks references, safe links, source provenance and deterministic output.

The journal lists actual research footprints and distinguishes source reviews, proposed protocols and executed fixtures. DVIDIA is the publisher and `hello@dvidia.org` is the team contact. Individual contributor attribution is intentionally empty pending confirmation; `@IAMMRRIVR` is a general follow link, not an asserted author identity. Raw authenticated research sessions, environment files and private execution paths are excluded from the public archive.

## Read claims at their actual scope

Expand Down
Loading
Loading