Isaac Sim — AV4EV Autonomous Go-Kart

Simulation environment for the AV4EV autonomous electric go-kart on the Purdue Grand Prix track using NVIDIA Isaac Sim.

Sensors

Sensor	Type	Status	ROS2 Topic
Livox HAP	LiDAR (RTX)	Working	livox/lidar
OAK-D RGB	Camera	Working	oak/rgb/image_raw
OAK-D Stereo L/R	Camera (stereo)	Working	oak/left/image_rect, oak/right/image_rect
OAK-D Depth	Camera (depth)	Working	oak/stereo/depth
ZED 2 Stereo L/R	Camera (stereo)	Working	zed2/zed_node/left/image_rect_color, zed2/zed_node/right/image_rect_color
ZED 2 Depth	Camera (depth)	Working	zed2/zed_node/depth/depth_registered
Point One Nav Atlas	GPS (via odom)	Bridge script	atlas/odometry → atlas/fix

Directory Structure

├── kart/                   # Go-kart 3D models (FBX + USD)
├── track/                  # Purdue track models + textures
├── stage/                  # USD scene files
│   ├── progress3 (copy).usd   # ← Active working scene
│   ├── progress3.usd          # Previous checkpoint
│   ├── progress2.usd
│   ├── progress1.usd
│   ├── Lidarprogress1.usd
│   └── rew.usd
└── scripts/
    ├── steering.py         # Vehicle control (WIP)
    ├── gps_bridge.py       # ROS2: odom → NavSatFix converter
    ├── patch_sensors.py    # USD patcher: LiDAR/camera config fixes
    ├── patch_cameras.py    # USD patcher: reparent cameras to Body
    ├── patch_lidar_fov.py  # USD patcher: LiDAR FOV 120°, reparent, lower heights
    └── revert_lidar.py     # USD patcher: revert LiDAR to ROTARY

Scene Hierarchy (progress3 (copy).usd)

/World
├── PhysicsScene
├── GroundPlane
├── kart [ArticulationRoot]
│   ├── Body [RigidBody]
│   │   ├── Cube_043 (chassis mesh)
│   │   ├── FL/FR/RL/RR (wheel joints)
│   │   ├── FL_steering/FR_steering
│   │   ├── OAK_D_RGB, OAK_D_Left, OAK_D_Right, OAK_D_Depth
│   │   ├── ZED2_Left, ZED2_Right, ZED2_Depth
│   │   └── (LiDAR attached here in GUI)
│   ├── wheel_FL/FR/RL/RR [RigidBody]
│   ├── FL_knuckle, FR_knuckle [RigidBody]
│   └── World_LivoxHAP [OmniLidar]
├── purdue_track
├── drive [OmniGraph — Ackermann keyboard controller]
├── lidar [OmniGraph — RTX LiDAR → ROS2]
├── CameraGraph [OmniGraph — 7 cameras → ROS2]
└── GPSGraph [OmniGraph — odometry → ROS2]

Running

Isaac Sim (GUI)

cd ~/isaac-sim
./isaac-sim.sh
# Open: ~/Downloads/isaac_sim/stage/progress3 (copy).usd
# Press ▶ Play to start simulation

GPS Bridge (separate terminal)

# Requires ROS2 Humble/Jazzy
python3 ~/Downloads/isaac_sim/scripts/gps_bridge.py

Configurable via ROS2 params:

• origin_lat (default: 40.4432 — Purdue track)
• origin_lon (default: -86.9427)
• origin_alt (default: 190.0m)

Verify ROS2 Topics

ros2 topic list
ros2 topic echo /livox/lidar        # LiDAR point cloud
ros2 topic echo /oak/rgb/image_raw  # OAK-D RGB
ros2 topic echo /atlas/odometry     # Odometry
ros2 topic echo /atlas/fix          # GPS NavSatFix (requires gps_bridge.py)

Known Issues

• LiDAR config: Livox HAP solid-state config (45,200 emitters) crashes the RTX sensor plugin due to empty emitter arrays in the config JSON. Using ROTARY/128-emitter with 120° FOV limit as workaround.
• GPS: No native Isaac Sim NavSatFix publisher — uses external ROS2 bridge script (gps_bridge.py).

Hardware Reference

Sensor	Model	Key Specs
LiDAR	Livox HAP	Solid-state, 120°×25° FOV, 150m range, 10Hz
Camera 1	OAK-D	RGB (IMX378) + Stereo (OV9282), 75mm baseline
Camera 2	ZED 2	Stereo 2208×1242, 110° HFOV, 120mm baseline
GPS	Point One Nav Atlas	RTK GPS, ~2cm accuracy