Overview of the code

The LIPM walking controller is built with the mc_rtc framework.

In mc_rtc, the role of a controller is to initialize and update a set of tasks for whole-body control. These tasks are in turn used to formulate and solve an inverse-kinematics or inverse-dynamics quadratic program (QP) using the Tasks library. Solving the QP provides joint torques that are sent directly to torque-controlled robots, or joint accelerations that are integrated into positions and sent to position-controlled robots. All of this is done automatically by the framework, so that the only task of a controller is to update... tasks.

There are two main functions in the controller:

• Controller::Controller initializes tasks and internal objects such as, in this controller, the stabilizer Stabilizer and walking pattern generator ModelPredictiveControl
• Controller::run is the main function which is called at every control cycle. When this function returns, all tasks should have been updated properly.

Main components

The Controller class has some main attributes that you will see everywhere in the code:

• Controller::controlRobot(): robot state output by the QP-based whole-body controller
• Controller::mpc(): model-predictive-control problem solver used for walking pattern generation
• Controller::pendulum(): holds the reference state (CoM position, velocity, ZMP, ...) from the walking pattern
• Controller::preview: current solution trajectory from the walking pattern generator
• Controller::plan: current footstep plan
• Controller::realRobot(): estimated robot state
• Controller::stabilizer(): balance feedback control component that updates task targets

As a general rule, look out for documentation in header files or on the API documentation.

Walking state machine

The last instruction in the controller's run() function calls mc_control::fsm::Controller::run(), which runs the main function of the current state in the finite state machine (FSM). Check out how the FSM works in the FSM wiki page. The four states in the walking controller's FSM are described in the Graphical user interface guide:

• states::Initial: nothing happens yet and the robot just holds its posture.
• states::Standing: the user clicked on "Start standing", the stabilizer is now on.
• states::SingleSupport: the robot is walking, currently in single support phase.
• states::DoubleSupport: the robot is walking, currently in double support phase.

If you just want some balance control without walking, you can modify the Standing state to suit your needs.

If you want the robot to do something while walking, you can extend the FSM and execute your new states in parallel with states::SingleSupport and states::DoubleSupport.