Genesis Simulation
Build and control robots with Genesis 1.3 on AMD GPUs, then advance from parallel simulation to ROCm perception, tactile interaction, and a guarded language-guided agent.
Goals
- Create Genesis scenes and load a Franka Panda robot
- Tune PD controllers and inspect robot state
- Solve inverse kinematics and execute pick-and-place motion
- Scale one scene into batched GPU environments
- Process vision and fingertip tactile observations with ROCm
- Validate language plans before executing robot actions
Genesis Labs
What this section covers
A six-lab progression from simulation fundamentals to an interactive, multimodal Physical AI agent.
GS01 — Hello Genesis
Initialize Genesis, create a scene, load a robot, and render camera observations.
GS02 — Control Your Robot
Apply PD control, tune gains, and command robot joints under gravity.
GS03 — Motion Planning
Use inverse kinematics and path planning to reach, grasp, and lift a cube.
GS04 — Parallel Simulation
Run multiple Franka environments in parallel for high-throughput experimentation.
GS05 — Perception with ROCm
Combine RGB, depth, segmentation, surface normals, and 8×8 fingertip tactile sensing in an interactive live HUD.
GS06 — Language-Guided Agent
Turn constrained language plans into perception-checked, tactile-gated robot actions with reproducible scene layouts.
Return to the Physical AI overview.