distributed-framework

D-VLA proposes a high-concurrency distributed asynchronous reinforcement learning framework for Vision-Language-Action models, using plane decoupling and a swimlane pipeline to improve throughput and efficiency in large-scale embodied AI training.