PiZero<T>

Models & Types · AiDotNet.VisionLanguage.Robotics

pi-zero: PaliGemma VLM with action expert for 8 robot embodiments.

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

Implements: IVisionLanguageAction<T>

For Beginners

pi-zero is a vision-language-action model for general robot control across multiple robot platforms. Default values follow the original paper settings.

How It Works

pi-zero (Black et al., 2024) is a vision-language-action flow model built on PaliGemma VLM with a dedicated action expert for general robot control. It uses flow matching to generate continuous robot actions, trained across 8 different robot embodiments for dexterous manipulation including folding, assembly, and bin-picking tasks.

References:

  • Paper: “pi0: A Vision-Language-Action Flow Model for General Robot Control (Black et al., 2024)”

Example

// Create a pi-zero model for general robot control
// PaliGemma VLM with action flow expert across 8 robot embodiments
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 PiZero<double>(architecture, "pizero.onnx");
            
// Training mode with native layers
var trainModel = new PiZero<double>(architecture, new PiZeroOptions());

Constructors

Constructor Summary
PiZero(NeuralNetworkArchitecture<T> architecture, PiZeroOptions options, IGradientBasedOptimizer<T, Tensor<T>, Tensor<T>> optimizer)
PiZero(NeuralNetworkArchitecture<T> architecture, string modelPath, PiZeroOptions 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 pi-zero’s PaliGemma VLM backbone + flow matching conditioning.
Tensor<T> PredictAction(Tensor<T> observation, string instruction) Predicts action using pi-zero’s flow matching formulation.