# Stanford Seminar - Distributional Representations and Scalable Simulations for Real-to-Sim-to-Real | Summary and Q&A

3.6K views
May 7, 2022
by
Stanford Online
Stanford Seminar - Distributional Representations and Scalable Simulations for Real-to-Sim-to-Real

## TL;DR

Addressing the difficulties in representing and simulating deformable objects using distributional embedding and differentiable simulation.

## Install to Summarize YouTube Videos and Get Transcripts

### Q: Why is it difficult to represent the state of deformable objects?

Deformable objects have an infinite number of degrees of freedom, unlike rigid objects with a fixed number of degrees of freedom. This poses a challenge in determining the state of a deformable object accurately.

### Q: What are the challenges in annotating the states of deformable objects?

Annotating the states of deformable objects is expensive and challenging. Placing markers on the objects for tracking purposes is a meticulous process, and using expensive scanners is not always feasible.

### Q: How do occlusions affect the representation of deformable objects?

Occlusions are challenging to handle in deformable objects, as their shape can change drastically within a short timeframe. Establishing the shape of an occluded deformable object becomes much more difficult due to this constant change.

### Q: How does reducing dimensionality impact the representation of deformable objects?

Mapping the representation of a deformable object into a lower dimensional space, even by tracking key points, does not eliminate the complex dynamics of the object. In the lower dimensional space, the key points jump around seemingly randomly, making it difficult to interpret the object's actual behavior.

## Summary & Key Takeaways

• Deformable objects have an infinite number of degrees of freedom, making it challenging to represent their states accurately.

• Annotation of deformable object states is expensive and difficult, requiring sophisticated scanning equipment.

• Dealing with occlusions, reducing dimensionality, and handling noisy states are additional challenges in representing deformable objects.