/TrajOpt

Trajectory optimization based on Taichi

Primary LanguagePython

TrajOpt

This is a trajectory optimization project (still updating). Implementation is based on Taichi 0.7.21.

Methods

  • Objective: reach a given target position
    • regularization
    • smooth trajectory (velocity constrain)
    • smoothness
  • Control parameters: forces per frame per node
  • Optimization method
    • gradient descent with line-search
    • Step and projection
    • L-BFGS
    • Gauss-Newton
  • Forward simulation
    • XPBD
    • Newton's Method
  • Backward computation: Adjoint Method

Experiments

Usage

  • asset/input.json sets initial and target position (.obj) and fixed points
  • In main.py
    • Set b_display to False to start optimization and save results
    • Set b_display to True to display forward simulation