## rbyte

Simplified dataset configuration for large, multimodal models from more than 50 popular spatial intelligence datasets.

### Features:
• ROS2/MCAP support  
• Inter-modality alignment  
• Dataframes with pola.rs  
• Visualization with rerun.io

[Go to Github](https://github.com/yaak-ai/rbyte)

## Tron

Train spatial intelligence leveraging our cutting-edge research on large, multimodal foundation models.

### Features:
• Episode construction and model configuration with hydra  
• Built with PyTorch Lightining  
• Pre-trained models and notebooks  
• Model tracking with wandb.ai  
• Inference visualization with rerun.io

## compose intelligence

### Align Modalities

Camera. Radar. Lidar. Audio. Controls. Easily align your multimodal data. rbyte supports conversion to mcap; a flexible format for all things spatial intelligence.

### Encode modalities

Chose from pre-trained modality encoders, or train modality encoders from scratch. Take inspiration from our research, and that of the wider AI community.

### Construct episodes

Easily construct training episodes from aligned, multimodal data through hydra.

### Train spatial intelligence

Leverage our cutting-edge research to train spatial intelligence foundation models.

#### Developer preview

Join the rbyte and Tron developer preview, and help us build spatial intelligence in the open.

[i'm interested](https://www.yaak.ai/contact)