The training and simulation process of UrbanDriver seem to be not using the kinematic state transition model. Is this right?

woven-planet / l5kit

L5Kit - https://woven.toyota

https://woven-planet.github.io/l5kit

851 stars 277 forks source link

The training and simulation process of UrbanDriver seem to be not using the kinematic state transition model. Is this right? #385

Open shubaozhang opened 2 years ago

shubaozhang commented 2 years ago

This is the unroll process. It seems that to produce the next state, it doesn't use the kinematic state transition model, but instead of using the predicted position as the next state directly. Is this true?