From Prompt to Walking Robot With NVIDIA RTX Spark
An NVIDIA presenter explains how RTX Spark, AI models, and simulation tools can take Micro Duck from a described course to a trained walking robot.
In this talk
- 0:08
Plan the robot’s course
The presenter introduces Micro Duck, a Hugging Face robot, and proposes teaching it vision and walking with NVIDIA RTX Spark. Hermes, Nemotron, and GPT-6 Astra help plan the task, with larger workloads sent to the cloud when needed.
- 0:52
Build a simulated world
Astra uses FreeCAD to create 3D objects and Blender to add materials, then draws on Omniverse SimReady libraries for physical properties. Isaac Sim brings those assets together into a scene for Micro Duck.
- 1:13
Model movement and train
Astra rigs the robot’s joints, actuators, and links, then simulates its movement with the Newton physics engine. Isaac Lab trains many robot instances in parallel in the cloud, refining the policy through repeated falls and lessons.
- 1:48
Add vision and deploy
Simulated cameras help Micro Duck learn to perceive its surroundings alongside its physics training. The presenter says the resulting robot can avoid obstacles, move through tight spaces, and adapt; the policy is then evaluated in Isaac Sim with multiple robots before deployment to the physical duck.
Briefs summarise talks and interviews; timestamps open the video at that moment. Claims in a talk are the speaker's own.