We present a framework that acts as an “intuitive physics reasoner” which takes in strategies expressed in natural language (whether from a human or LLM), and assesses their validity based on a physics knowledge library. We believe the ability to quickly determine whether a strategy is worth considering and allocating further resources to planning using that strategy is useful. The reasoner can verify that a strategy is consistent with its prior knowledge of physics, and that inputs, states, and outputs are within set limits.
It is also capable of finding optimal parameters according to a given criterion.

We will show examples of physical problems involving general physics (including liquids) and thermodynamics.

Chris Atkeson

David Held

Oliver Kroemer

Leonid Keselman