This model is interesting, it looks like it uses ML. I wonder if you would need to have a differently trained model for every animation space you want to use (e.g. pitching baseball, running, rolling on the ground, doing the limbo)
aishikbhattacharyya
There's also a form of kinematics known as FK, or forward kinematics, which focuses on the joints as the main predictor for where the body part is.
J1ungle
In my understanding, each joint of the body can only rotate within a limited range. With the hierarchical representation of the human body, could we potentially filter out body gestures that exceed these constraints? I believe this might offer a more cost-effective solution compared to the method outlined in this slide.
sebzhao
This type of stuff is really cool! I know imitation learning is used in some ML research order to get the kinematics correct. Interesting to see what technique ultimately prevails.
This model is interesting, it looks like it uses ML. I wonder if you would need to have a differently trained model for every animation space you want to use (e.g. pitching baseball, running, rolling on the ground, doing the limbo)
There's also a form of kinematics known as FK, or forward kinematics, which focuses on the joints as the main predictor for where the body part is.
In my understanding, each joint of the body can only rotate within a limited range. With the hierarchical representation of the human body, could we potentially filter out body gestures that exceed these constraints? I believe this might offer a more cost-effective solution compared to the method outlined in this slide.
This type of stuff is really cool! I know imitation learning is used in some ML research order to get the kinematics correct. Interesting to see what technique ultimately prevails.