GRID

Run high-fidelity simulations with a variety of simulators such as AirGen and Isaac Sim. Import your own robots, or entire 3D scenes through Gaussian splatting. Leverage Unreal Engine 5 or Omniverse for scalable and performant data generation, policy training, and domain randomization - either in the cloud or on-prem.

"sim": {
        "sim_type": "airgen",
        "scene_name": "gsplat",
        "kwargs": {
            "geo": false,
            "ply_file": "abandoned_mine.ply"
        }
}

Generate domain-rich datasets at scale for your form factor of choice - by bringing in realistic objects, equipping a variety of sensors, and varying weather and other environmental conditions. Parallelize over multiple machines, capturing camera, LiDAR, as well as rich ground truth for comprehensive robot training data.

def readSensors(client) -> dict:
    sensor_data["imu"] = client.getImuData()
    sensor_data['rgb'] = client.getImages("front_center",[airgen.ImageType.Scene])[0]
    sensor_data["lidar"] = client.getLidarData()
    return sensor_data

@data_collector(readSensors, time_delta=0.1)
def move_task(client, trajectory):
    client.moveOnPath(trajectory)

Invoke state-of-the-art AI models with just a few lines of code to create complex skills for your robots. Train or fine-tune your own, leveraging built-in autonomy stacks for mapping, planning, and trajectory generation. Chain multiple models for complex tasks, track performance across a variety of scenarios, and streamline iteration until your solution is fully deployment-ready.

from grid.model.perception.vlm.moondream import MoonDream 
from grid.model.perception.segmentation.oneformer import OneFormer 
from grid.model.perception.depth.metric3d import Metric3D 

depth_model = Metric3D()
seg_model = OneFormer()
vl_model = MoonDream() 

depth_pred = depth_model.run(rgb_image.data)
seg_pred = seg_model.run(rgb_image.data, mode="panoptic")
vl_model.run(rgb_image.data, "Describe what you see")

Seamlessly deploy into robots with native support for industry standard protocols like ROS2, MAVLINK, and more. Efficient middleware techniques and cloud-centric workflows enable dynamic orchestration and effortless scalability.