星尘智能发布 SmoothRL，让在线强化学习跟上大模型的异步推理

As VLAs and World-Action Models grow larger and inference times stretch longer, robots in real-world deployments have long since switched to asynchronous inference — executing task A while the model computes task B in th…