EnvHub

EnvHub is Nepher Robotics' Isaac Lab environments platform — a versioned library of OpenUSD simulation scenes used for fair tournament training, standardized evaluation, and community reuse.

Browse and download environments at nepher.ai/envhub, or from the terminal with npcli envhub.

Why EnvHub Exists

Tournament integrity depends on everyone — miners and validators — running against the same environments. EnvHub centralizes that:

• Standardized bundles — navigation, manipulation, humanoid, and locomotion categories
• Versioned releases — semantic versions with stable IDs (e.g. waypoint-benchmark-v1, waypoint-sample-v1)
• Reproducible evaluation — validators download the exact scenes miners train against
• OpenUSD assets — Omniverse-native scenes compatible with Isaac Sim 5.1+ / Isaac Lab 2.3+

Without EnvHub, benchmarks drift: different mesh versions, hidden scene tweaks, or unreproducible layouts undermine trust in leaderboard results.

How Environments Are Structured

Each environment is a ZIP bundle identified by a unique ID (e.g. waypoint-benchmark-v1) and organized by category (navigation, manipulation, humanoid, etc.).

Every bundle contains:

• manifest.yaml — required metadata describing scenes, version, and configuration
• Scene files — OpenUSD (.usd) scenes or programmatic preset scenes generated from Python configs

A single environment can hold multiple scenes — variations of terrain, obstacles, or waypoint layouts within the same theme. Scenes are what Isaac Lab actually loads during training or evaluation.

Scene Types

Type	Description
USD	Static 3D environments loaded from OpenUSD files; may include occupancy maps
Preset	Dynamically generated scenes from Python configuration with domain randomization

Benchmarks & Evaluation

Some environments are marked as benchmarks — standardized test sets used for tournament evaluation.

• Miners download training/sample environments (e.g. waypoint-sample-v1) during the contest period
• Benchmark environments (e.g. waypoint-benchmark-v1) are reserved for validators and only activated during the evaluation period — preventing overfitting to test scenes
• Validators query active evaluation benchmarks and download them when scoring begins
• Active evaluation benchmarks are listed publicly (without their contents) at nepher.ai/envhub/eval-benchmarks

This is the mechanism that makes Tournament leaderboards fair and reproducible.

Using EnvHub as a Miner

Download environments before training or self-evaluation:

pip install nepher-cli
npcli login --api-key nepher_xxxxxxxx

# List available environments
npcli envhub list --category navigation

# Download training / sample scenes for a navigation tournament
npcli envhub download waypoint-sample-v1

# Inspect local cache (default ~/.nepher/cache)
npcli envhub cache list

Then train with the EnvHub-integrated task variant in your tournament repo:

# Example: Spot waypoint navigation (Isaac Lab 2.3.2 + Isaac Sim 5.1)
python scripts/rsl_rl/train.py --task=Nepher-Spot-WaypointNav-Envhub-v0

See task-spot-waypointnav for a complete integration example and the Miner Guide for the full workflow.

Using EnvHub as a Validator

The Subnet 49 validator downloads the active evaluation benchmarks automatically with its Nepher API key when the evaluation period starts — no manual step is needed. Bundles are cached in data/env-cache inside the nepher-subnet directory.

Evaluation runs via eval-nav with env_scenes pointing to downloaded EnvHub IDs. See the Validator Guide.

Bundle Categories

Category	Use case
Navigation	Waypoint following, obstacle avoidance, indoor/outdoor terrains
Manipulation	Pick-and-place, articulated object interaction
Humanoid	Locomotion and whole-body tasks
Locomotion	Legged and wheeled mobility benchmarks

New categories are added as tournament tasks require them.

Contributing Environments

Creators can upload bundles (subject to admin review):

npcli envhub upload ./my-bundle --category navigation

Submission guidelines:

• OpenUSD scene files compatible with Isaac Sim 5.1+
• Valid manifest.yaml describing all scenes
• Compliant Python / Isaac Lab integration where applicable
• Clear documentation and preview media
• Passed QA for collision meshes, scale, and waypoint validity

You can also upload from the web at nepher.ai/envhub/upload and track status under My Uploads.

Uploads enter a pending → approved/rejected workflow. Approved environments may appear in future tournaments or on SimStore.

Access & Quotas

Sign in with your Nepher account, or authenticate the CLI with an API key from nepher.ai/account/api-keys.

Role	Typical access
Miner / user	Approved non-benchmark environments + own uploads
Validator	Active evaluation benchmarks during tournament eval periods
Admin	Full approval workflow and benchmark activation

Default limits:

Limit	Default
Maximum bundle size	10 GB
Storage per user	500 GB (admins 1 TB)
Uploads per day	100
Download bandwidth	100 GB per 30-day window

Your current usage is shown on your EnvHub dashboard. Contact the team on Discord if a workload needs higher limits.

Recommended interface: Nepher CLI (npcli envhub).

Related

• Tournaments — competitions that consume EnvHub benchmarks
• NVIDIA Omniverse — OpenUSD and Isaac Lab foundation
• Miner Guide — full workflow including EnvHub setup
• eval-nav — evaluation framework referencing EnvHub scenes