F3 · Robotics & Embodied
Isaac Sim Robotics on DGX Spark
Overview
On an Arm-based NVIDIA DGX Spark, this guide builds Isaac Sim and Isaac Lab from source, runs Cartpole and articulated-robot simulation, and trains and evaluates a Unitree H1 humanoid locomotion policy with RSL-RL PPO. The Grace CPU handles environment coordination, reward calculation, sensor preprocessing and software build, while the Blackwell GPU runs physics, rendering, parallel simulation and policy training. This Arm64 robotics baseline chains simulation, reinforcement learning and the heterogeneous compute platform into a full path from environment validation to humanoid-policy evaluation.
Scope of verification
- It is an Isaac Sim / Isaac Lab robotics simulation and RL reference workflow on a single DGX Spark, validating the simulation environment, RSL-RL PPO training and Unitree H1 policy evaluation.
- Dependencies include the Grace Blackwell GB10, compatible Isaac Sim/Isaac Lab versions, RSL-RL, task configuration, GPU drivers, a Python environment and sufficient system/GPU memory.
- Scope centers on simulation and policy training; it provides no CPU/GPU utilization, training speedup, latency or cross-platform comparison, and the policy is not deployed to a physical Unitree H1, so sim-to-real, robot safety and hardware-control validation are not covered.
Canonical source: https://learn.arm.com/learning-paths/laptops-and-desktops/dgx_spark_isaac_robotics/
This page is a guided summary with boundary notes, not an official reproduction.
This page is a guided summary with boundary notes, not an official reproduction.