sonic

Manages NVIDIA GEAR-SONIC whole-body-control model lifecycle including training, export, evaluation, and serving across GPU clusters.

17|8|Updated Apr 7, 2026
One-click install
npx skills add https://github.com/nebius/nebius-physical-ai --skill sonic
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: sonic
Source: https://github.com/nebius/nebius-physical-ai/tree/main/skills/tools/sonic
Command: npx skills add https://github.com/nebius/nebius-physical-ai --skill sonic

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

This skill streamlines the complex lifecycle of NVIDIA GEAR-SONIC whole-body-control models, removing the friction of managing disparate training, export, and deployment tasks.

Core Features & Use Cases

  • Unified Lifecycle Management: Orchestrate training, evaluation, and serving through a single command surface.
  • Infrastructure Routing: Automatically handle image routing and GPU target validation for diverse environments like L40S VMs and Blackwell Kubernetes clusters.
  • Use Case: A robotics researcher can use this skill to train a locomotion policy, export the artifact, and deploy it to a Kubernetes-based GPU cluster without manually configuring environment-specific image variants or S3 paths.

Quick Start

Execute the npa workbench sonic train command to initiate a new whole-body-control policy training session on the configured Nebius infrastructure.

Frequently Asked Questions about sonic

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I manage end-to-end NVIDIA whole-body-control robot policy training and deployment?

You can manage the full lifecycle of NVIDIA GEAR-SONIC whole-body-control models by orchestrating training, evaluation, and serving through a single command surface. This provides unified lifecycle management for complex robotics workflows.

Can I deploy whole-body-control models to heterogeneous GPU targets like L40S and Blackwell Kubernetes?

Yes, whole-body-control deployment supports heterogeneous GPU targets such as L40S VMs and Blackwell Kubernetes clusters. The system automatically handles image routing and validates GPU targets to ensure correct environment configuration.

How does artifact routing and validation work for robotics model export?

Artifact routing and validation for robotics model export works through integrated image manifest handling and S3 path scoping. This ensures consistent artifact delivery across diverse deployment environments without manual path configuration.

What is the best way to start a whole-body-control locomotion policy training session?

The best way to start a whole-body-control locomotion policy training session is to execute the npa workbench sonic train command. This initiates training on the configured Nebius infrastructure without manual environment setup.

Do I need to manually configure environment-specific image variants for GPU cluster serving?

No, you do not need to manually configure environment-specific image variants for GPU cluster serving. The infrastructure routing feature automatically handles image routing and GPU target validation for diverse environments.

Why does consistent artifact routing matter for whole-body-control model evaluation?

Consistent artifact routing matters for whole-body-control model evaluation because it ensures validated artifacts are correctly delivered across heterogeneous environments. Integrated S3 path scoping prevents misalignment during the export and serving phases.