Tutorial - OCS2 - Farbod Farshidian

OCS2 is a C++ toolbox tailored for Optimal Control of Switched Systems (OCS2). The toolbox provides an efficient implementation of the Differential Dynamic Programming (DDP) algorithm in continuous-time (known as SLQ) and discrete-time (known as iLQR) domains. OCS2 handles general path constraints through augmented Lagrangian and relaxed barrier methods. To facilitate the application of OCS2 in robotic tasks, it provides the user with additional tools to set up the system dynamics (such as kinematic or dynamic models) and cost/constraints (such as self-collision avoidance and end-effector tracking) from a URDF model. The library also provides an automatic differentiation tool to calculate derivatives of the system dynamics, constraints, and cost. The toolbox’s efficient and numerically stable implementations in conjunction with its user-friendly interface have paved the way for deploying it in an MPC fashion for numerous robotic applications with limited onboard computation power.