Skip to content

Hierarchy in wiki:

---

Backward methods

Auto Diff

语言层面的自动微分计算。

Explicit method.

  • 计算图（PyTorch）
  • Tape（Tensorflow、Taichi）

Papers:

  1. GradSim (∇Sim: DIFFERENTIABLE SIMULATION FOR SYSTEM IDENTIFICATION AND VISUOMOTOR CONTROL)
  2. A Differentiable Physics Engine for Deep Learning in Robotics
  3. Taichi
  4. ChainQueen: Cuda

Dual Diff

和前向仿真同时计算，使用 Taylor 展开

  1. DiffFR

Adjoint Method

通常基于 Implicit method

  1. DiffPD, DiffCloth
  2. DiffXPBD
  3. ADD
  4. Differentiable Fluids with Solid Coupling for Learning and Control
  5. Efficient Differentiable Simulation of Articulated Bodies(2021 icml)
  6. Fluid Control Using the Adjoint Method (2004)
  7. Keyframe Control of Complex Particle Systems Using the Adjoint Method