AI RESEARCH
OxyGen: Unified KV Cache Management for Vision-Language-Action Models under Multi-Task Parallelism
arXiv CS.AI
•
ArXi:2603.14371v1 Announce Type: cross Embodied AI agents increasingly require parallel execution of multiple tasks, such as manipulation, conversation, and memory construction, from shared observations under distinct time constraints. Recent Mixture-of-Transformers (MoT) Vision-Language-Action Models (VLAs) architecturally such heterogeneous outputs, yet existing inference systems fail to achieve efficient multi-task parallelism for on-device deployment due to redundant computation and resource contention. We identify isolated KV cache management as the root cause.