Open research questions

From Bob Wiki, a public encyclopedia-style profile page

These are public research questions Bob expects to keep returning to. They help convert a broad long-term vision into concrete research problems.

1. How can animals be reconstructed in 4D without dense supervision?
2. What physical priors are useful for non-human articulated bodies?
3. Can human motion models transfer to animal motion and human-animal interaction?
4. What representations best support animatable, controllable animal avatars?
5. How can 3D/4D reconstruction become a body layer for AI agents?
6. How can neural rendering, physical simulation, and control be unified for embodied characters?
7. What is the smallest convincing benchmark for general animal 4D reconstruction?

Visiting research framing

A concrete current framing is physics-grounded 4D animal reconstruction from in-the-wild videos. This connects Bob’s animal reconstruction background with neural representations, dynamic reconstruction, physically plausible motion, contact, locomotion, and embodied character modeling.

Questions for motion / embodied AI labs

1. How can reconstruction systems expose motion states that are useful for downstream simulation or control?
2. Which physical priors matter most for believable animal and character motion: contact, balance, energy, kinematic limits, or learned motion manifolds?
3. Can synthetic animal video pipelines be used not only for reconstruction training, but also for evaluating physical plausibility and controllability?

Needs prioritization