Tournaments

Tournaments are rigorous, focused competitions centered on benchmarking and policy improvement. They are the primary engine through which Nepher Robotics produces valuable AI assets — not just leaderboard rankings.

Browse live and past tournaments at nepher.ai/tournament.

What You Get From a Tournament

Every tournament is designed to leave behind reusable robotics IP:

Deliverable	Details
Policy checkpoints	Trained weights (best_policy.pt) from the approved podium
Isaac Lab task code	Full External Project repos with training and play scripts
Evaluation configs	Public scoring via eval-nav
Environment assets	OpenUSD scenes from EnvHub used in evaluation
Highlights	Verified play videos of ranked runs at nepher.ai/tournament/highlights
Documentation	Task-specific READMEs (e.g. Spot Waypoint Navigation)

After review, podium agents are made available according to the tournament's winner access mode — private, public (free download for signed-in users), or paid unlock. Selected winning packages are also listed on SimStore by the Nepher team.

Purpose & Scope

• Purpose: Drive optimization and identify the highest-performing robotic control policies for a well-defined task.
• Scope: Fixed robot, environment, and metrics — e.g., "Boston Dynamics Spot waypoint navigation across EnvHub benchmark scenes".
• Metrics: Objective, task-specific measurements defined in the tournament's eval-nav config (for navigation: success rate and time-to-completion).
• Stack: Isaac Sim 5.1, Isaac Lab 2.3.2, RSL-RL training, EnvHub OpenUSD scenes.
• Validation: Multiple independent validators score each agent using the same public eval-nav pipeline on Decentralized Physical AI infrastructure (Bittensor Subnet 49).
• Rewards: The approved podium (up to 3 places, default split 70% / 20% / 10%) shares the subnet's miner emissions during the reward period. See Incentive Mechanism.

What Competitors Optimize

Participants compete on the full stack of robot learning:

• RL algorithm design (reward functions, weights, auxiliary objectives)
• Training methodology (curriculum, domain randomization, hyperparameters)
• Data and environment selection (EnvHub scene diversity — train on waypoint-sample-v1, evaluated on hidden waypoint-benchmark-v1 scenes)
• Model architecture within the allowed framework (e.g., RSL-RL actor-critic)

Example Task Repository

The task-spot-waypointnav repo is a reference tournament task:

• Robot: Boston Dynamics Spot quadruped
• Architecture: Hierarchical — trainable high-level nav policy (3-dim velocity commands) + frozen low-level locomotion policy
• Requirements: Isaac Lab 2.3.2, Isaac Sim 5.1
• Registered envs: Nepher-Spot-WaypointNav-v0, Nepher-Spot-WaypointNav-Envhub-v0, and Play variants
• EnvHub integration: Standardized navigation scenes via npcli envhub download
• Self-evaluation: Same eval-nav configs validators use

Clone it as your starting point, or follow the structure when a new tournament task is announced.

Participation

When a tournament opens:

1. The task GitHub repo is published (environment, robot USD/URDF, training scripts)
2. Required EnvHub environments are listed for download via Nepher CLI
3. The evaluation config is defined in eval-nav YAML format
4. The tournament page shows the schedule, rules, reward allocation, and leaderboards

Multiple tournaments can run at the same time; their reward periods are scheduled so they do not overlap. Task repos follow the Isaac Lab External Project pattern — the standard approach for custom robot policies on Omniverse.

To take part you need:

• A Nepher account and an API key
• A Bittensor hotkey registered on Subnet 49
• A registered coldkey and a funded platform balance to pay submission fees

Tournament Lifecycle

Period	Description
Contest	Miners train policies and upload signed agent archives. With public evaluation enabled, validators score submissions on public scenes and a public leaderboard updates live.
Submit	Final upload window. The eligibility snapshot (registered on Subnet 49 + submitted agent) is built before evaluation.
Evaluation	Validators run eval-nav on hidden EnvHub benchmark scenes in headless Isaac Sim and produce the private leaderboard.
Review	The admin team verifies the top submissions for integrity and approves the podium, or confirms no winner.
Reward	Validators assign on-chain weight to the podium; highlights and winner downloads are published.

Scores are stake-weighted across validators, ties go to the earliest upload, and only your latest clean upload counts. Full details are in the Incentive Mechanism.

Submission Fees

Each upload is charged a small fee from your platform balance. Your first upload in a tournament costs the minimum (default 0.1 TAO); right after each upload the next fee jumps to the maximum (default 0.5 TAO) and decays back to the minimum over about 6 hours. Your personal next fee is shown on nepher.ai/tournament/submission-cost. See Platform Balance & Fees.

Security Review

Every upload is automatically scanned for disallowed behavior (network calls, spawning processes, dynamic imports, reading secrets or wallets, banned file types). Flagged agents are not evaluated until an admin clears them; confirmed abuse leads to account blocking. Validators additionally run every agent inside an isolated GPU sandbox.

Notifications

Signed-in users receive emails when a contest starts, when the submit window opens, and when winners are announced. Manage them under Settings → Notifications on nepher.ai/tournament/settings/notifications.

Sponsored Tournaments

Enterprises can sponsor a tournament for brand visibility and recruitment. Sponsored events are listed at nepher.ai/tournament/sponsored. Sponsors never control hidden tests or scoring. See Sponsored Events.

Platform URLs

Service	URL
Tournaments	nepher.ai/tournament
Submission cost	nepher.ai/tournament/submission-cost
Highlights	nepher.ai/tournament/highlights
Account & API keys	nepher.ai/account/api-keys
Billing (platform balance)	nepher.ai/account/billing
EnvHub	nepher.ai/envhub
Tournament API	https://api.nepher.ai/tournament

Submit via CLI

pip install nepher-cli bittensor-wallet
npcli login --api-key nepher_xxxxxxxx
npcli account balance                      # make sure you can cover the submission fee
npcli tournament list-active
npcli tournament check --path ./my-agent
npcli tournament submit --path ./my-agent --wallet-name miner --wallet-hotkey default

Agent submissions must include best_policy/best_policy.pt and an Isaac Lab task module under source/<module>/. See Nepher CLI and the Miner Guide for the full workflow.

Guides & Tools

Resource	Link
Miner guide	docs.nepher.ai/miner-guide
Validator guide	docs.nepher.ai/validator-guide
Incentive mechanism	docs.nepher.ai/im
Nepher CLI	docs.nepher.ai/nepher-cli
GitHub org	github.com/nepher-ai