NWM is a multivariate autoregressive generative embodied driving model that can fully understand data and has long-term deduction and decision-making capabilities. It can deduce 216 possible trajectories in 0.1 seconds, find the best decision, and then repeatedly update the internal space-time model in the next 0.1 seconds based on the input of external information, and then predict 216 possibilities. In this cycle, continuous prediction follows the driving trajectory to obtain the optimal driving solution.
NWM一个多元自回归生成式的具身驾驶模型，可全量理解数据、具有长时序推演和决策能力。它可以在0.1秒内推演出216种可能发生的轨迹，寻找最佳决策，然后在接下来的0.1秒内，根据外界的信息的输入，重复更新内在时空的模型，再去预测216种可能性。以此循环，跟随驾驶轨迹持续预测，得到驾驶的最优解。