fine-tuning-openvla-oft

What problem does it solve?

This workflow provides end-to-end fine-tuning and evaluation of OpenVLA-OFT policies for robot action generation, enabling researchers to adapt LoRA-based models with FiLM conditioning for LIBERO and ALOHA pipelines.

Core Features & Use Cases

• Fine-tune OpenVLA-OFT with LoRA adapters and optional FiLM conditioning for improved grounding.
• Supports LIBERO-like simulation tasks and ALOHA real-world pipelines with multi-image inputs and proprioception.
• Reproducible deployment: merge LoRA adapters and deploy via VLA server for client evaluations.

Quick Start

Run the provided OpenVLA-OFT fine-tuning workflow with your dataset to start training and evaluation.