Add a New Object

This tutorial guides you through the process of adding a new object primarily using AI Hub. The steps include registering a 3D mesh model, preparing an object detection model, and configuring pose estimation and grasping capabilities. Follow these steps to enable your robot to recognize, locate, and manipulate new objects.

Step 1. Register a 3D Mesh Model

1.1 Via externally sourced mesh model

You can upload a pre-existing 3D mesh model of an object to the AI Hub using AI Hub: Data Management. See Tutorial: Upload A Mesh Model

See also

For a programmatic approach to perform data management, refer to API: Data Management Client.

1.2 Via 3D scanning

Use AI Hub: 3D Object Scanning (Scan Box only) to create a texturized 3D mesh model of your desired object(s).

See also

For a programmatic approach to perform workspace/environment scanning, refer to API: Robot Scan Client.

Step 2. Prepare an Object Detection model

2.1 Generate a Training Dataset

Use AI Hub: Data Generation to generate one/several dataset(s) consisting your desired object(s).

For detailed instructions on creating synthetic and/or real datasets, refer to:

Tutorial: Generate Synthetic Dataset

Tutorial: Collect Real Data

Tutorial: Annotate Real Data

See also

To programmatically generate a dataset, refer to the following client APIs:

2.2 Train an Object Detection Model

Utilize AI Hub: Model Training and Testing to train an object detection model on your registered dataset(s). See Tutorial: Train an Object Detection Model

See also

For a programmatic approach to train and infer an object detection model, go to API: Instance Segmentation Client

Step 3. Prepare a Pose Estimation model

The default pose estimation model general/v_1 that uses the method neura_nonDLPE1 is a deterministic pose estimation algorithm. After registering the object mesh model, no additional action is required to use this model to perform 6D object pose estimation.

Attention

Currently, no pose estimation method that require training are available.

See also

For a programmatic approach to infer a pose estimation model, go to API: Pose Estimation Client

Step 4. Enable Grasping

To perform pose-aware picks in an application, one is required to record grasps of the objects that will be picked. See Tutorial: Enable Grasping of New Object to learn how to teach grasp poses.