maple.backend.policy.openvla

OpenVLA policy backend.

This module implements the policy backend for OpenVLA (Open Vision-Language-Action), a vision-language-action model for robotic manipulation. OpenVLA takes visual observations and natural language instructions as input and outputs robot actions.

OpenVLA is based on transformer architectures and requires action unnormalization using dataset statistics to produce executable robot commands. The model is served via Docker containers with the inference API accessible over HTTP.

Available versions: - 7b: OpenVLA 7B parameter model - latest: Alias for the 7B model (default)

Classes

OpenVLAPolicy()

Backend for OpenVLA vision-language-action models.