Crocoddyl (Contact RObot COntrol by Differential DYnamic programming Library)

A contact-aware optimal control library that solves robot trajectories using differential dynamic programming.

The name is an acronym for Contact RObot COntrol by Differential DYnamic programming Library, open-sourced by researchers at the University of Edinburgh and LAAS-CNRS (Mastalli and others). It formulates optimal-control problems for multi-body dynamics with contact in a unified way, then iterates on solvers such as DDP and FDDP to produce joint torques and a full trajectory — fast enough to run as online model predictive control. Legged robots can't avoid this class of tool: jumping or climbing a step requires simultaneously deciding foothold locations, contact forces, and whole-body motion, which inverse kinematics alone can't produce. It uses Pinocchio for the underlying dynamics, provides C++ and Python interfaces, and is often compared with OCS2 and acados.

ExampleTo plan a jump for a quadruped robot, define the contact sequence and cost function in Crocoddyl and solve for the torque trajectory with FDDP.

Related

Differential Dynamic Programming · Optimal Control · Trajectory Optimization · Model Predictive Control · Pinocchio · OCS2

Sources

loco-3d/crocoddyl - GitHub