Manage simulation environment and evaluation

Browse simulation assets

1. In the left-side navigation pane, expand Simulation Management and click Simulation Assets.

2. Simulation assets are organized into three tabs:

   Tab	Description
   Robots	Robot models, such as various configurations of the Unitree G1 series.
   Environments	Simulation scenes, such as warehouses and laboratories.
   Objects	Interactive objects, such as tables and drawers.

The platform includes built-in official assets from Unitree and Kujiale, which can be installed and used directly. Each asset supports the Details operation and the Uninstall (for official assets) or Delete (for user assets) operation.

Upload custom assets

1. On the Simulation Assets page, click Upload Asset.

2. Configure the following parameters:

   Parameter	Required	Description
   Category	Required	Robots, Environments, or Objects.
   Asset Name	Required	The name of the asset. An asset folder is created with this name.
   File	Required	Supports .usd, .usda, .usdc, .usdz, or .zip (automatically decompressed) formats.

3. Click OK.

Create a simulation development environment

1. In the left-side navigation pane, click Simulation Development.

2. Click New simulation environment.

3. Configure the following parameters:

   Parameter	Required	Description
   Environment name	Required	The name of the simulation environment.
   Image	Required	Select an official simulation image or a custom image.
   Resource spec	Required	Set the Resource type to GPU, and then select a GPU specification.
   Service ports	Optional	Port 20001 (xpra visualization) is included by default.
   Environment Variables	Optional	Environment variables in key-value format.
   SSH public keys	Optional	The public key used for SSH logon.

4. Click Confirm.

Official simulation images

Differences from model development machines

Connect to an inference service

When running evaluations on a simulation development machine, you need to connect to a deployed inference service:

• Built-in model inference: Use the inference service URL generated by one-click deployment on the platform. For more information, see Train and deploy built-in models.

• Custom model inference: Manually start the inference service on the development machine. For more information, see Develop custom models.

Create a simulation image

After customizing the simulation environment, you can package it as a custom simulation image. The process is the same as creating an image for a model development machine. After pushing the image to ACR, it appears on the Image Management page of the platform.

Submit an evaluation task

1. In the left-side navigation pane, click Simulation Eval.

2. Click New evaluation.

3. Configure the following parameters:

   Parameter	Required	Description
   Task Name	Required	The name of the evaluation task.
   Inference service query	Optional	View the URL of a deployed inference service. You need to manually add the URL to the start command.
   Image	Required	Select a simulation environment image.
   Compute resources	Required	Select a GPU specification.
   Start command	Required	The command to start the evaluation script.
   Environment Variables	Optional	Environment variables in key-value format.
   Save as Template	Optional	Save the current configuration as an evaluation template.
   Load from Template	Optional	Select a saved evaluation template.

4. Click Confirm.

Monitor evaluations

After an evaluation task is submitted, you can view the following information in the evaluation list:

Download evaluation results

After the evaluation is completed, you can download the evaluation report and videos:

• Browser download: Suitable for files smaller than 200 MB.

• Export to OSS: Suitable for large files.

The evaluation report is typically in the eval_report.json format and contains evaluation metrics and result statistics.