EdgeVLA is a specialized Claude Code skill designed to streamline the development and deployment of Vision-Language-Action (VLA) models for robotics. By standardizing diverse datasets from the Open-X Embodiment (OXE) collection—including DROID, Bridge, and LIBERO—it provides a unified framework for training robots to interpret visual data and language instructions into physical actions. This skill is essential for robotics engineers and researchers who need consistent data formatting, multi-view camera support, and optimized inference capabilities for edge-based robot controllers.
主要功能
01Flexible Encoding: Supports POS_EULER, POS_QUAT, and JOINT state and action encodings.
02OXE Dataset Standardization: Maps diverse robotics datasets into a unified format for consistent training.
03Named Data Mixtures: Provides pre-configured mixtures like 'oxe_magic_soup' for robust model generalization.
04Multi-Camera Support: Handles primary, secondary, and wrist-mounted camera views simultaneously.
05Edge-Ready Inference: Optimized for low-latency performance on edge robotics hardware.
062 GitHub stars
使用场景
01Finetuning foundation VLA models on custom robot hardware and specific environments.
02Normalizing disparate Open-X Embodiment datasets for large-scale multi-dataset training.
03Implementing natural language command execution for humanoid and mobile robots at the edge.