ThreeDVLA<T>

Models & Types · AiDotNet.VisionLanguage.Robotics

3D-VLA: connects vision-language-action to 3D world via generative world model.

Inheritance: NeuralNetworkBase<T> → VisionLanguageModelBase<T> → ThreeDVLA<T>

Implements: IVisionLanguageAction<T>

For Beginners

3D-VLA is a vision-language-action model that uses 3D world modeling for robot planning and action generation. Default values follow the original paper settings.

How It Works

3D-VLA (UMass, 2024) connects vision-language-action models to the 3D world via a generative world model. It processes 3D point cloud observations alongside language instructions and generates both predicted future states (as a world model) and robot actions, enabling planning through imagination in 3D space.

References:

  • Paper: “3D-VLA: A 3D Vision-Language-Action Generative World Model (UMass, 2024)”

Example

// Create a 3D-VLA model for 3D vision-language-action reasoning
// connecting VLA to 3D world via a generative world model
var architecture = new NeuralNetworkArchitecture<double>(
    inputType: InputType.TwoDimensional,
    taskType: NeuralNetworkTaskType.ImageClassification,
    inputHeight: 224, inputWidth: 224, inputDepth: 3, outputSize: 512);
            
// ONNX inference mode with pre-trained model
var model = new ThreeDVLA<double>(architecture, "threedvla.onnx");
            
// Training mode with native layers
var trainModel = new ThreeDVLA<double>(architecture, new ThreeDVLAOptions());

Constructors

Constructor Summary
ThreeDVLA(NeuralNetworkArchitecture<T> architecture, ThreeDVLAOptions options, IGradientBasedOptimizer<T, Tensor<T>, Tensor<T>> optimizer)
ThreeDVLA(NeuralNetworkArchitecture<T> architecture, string modelPath, ThreeDVLAOptions options)

Properties

Property Summary
int ActionDimension { get; }
int DecoderEmbeddingDim { get; }
int EmbeddingDimension { get; }
int MaxGenerationLength { get; }
string LanguageModelName { get; }

Methods

Method Summary
ModelMetadata<T> GetModelMetadata()
Tensor<T> EncodeImage(Tensor<T> image)
Tensor<T> GenerateFromImage(Tensor<T> image, string prompt) Generates from image using 3D-VLA’s generative world model approach.
Tensor<T> PredictAction(Tensor<T> observation, string instruction) Predicts action using 3D-VLA’s generative world model approach.