Prototype differentiable simulator for robotic systems. This repository includes models for planar hopper, biped, and quadruped systems.
Systems are represented in minimal coordinates and contact impulses are computed at each time step by solving a nonlinear complementarity problem (NCP) using a custom interior-point solver. Gradients through the dynamics are efficiently computed via the implicit-function theorem applied to the NCP optimality conditions.
For more details, see our paper: Dojo: A Differentiable Simulator for Robotics and checkout our full simulator: Dojo.
RoboDojo can be added via the Julia package manager (type ]):
pkg> add RoboDojo
Please see the following for examples using this package: