1 Feed Forward
Within the realm of control systems feedforward usually refers to an open loop command provided to a closed loop system. In LaserCom this often means using an IMU to command a fast steering mirror to reduce the residual error by predicting where the target is relative to the instantaneous position of the spacecraft. This site also provides a feedforward example of a hovering helicopter.
1.1 Feedback Aspect
Typically there an actuator that is being commanded by raw, filtered, or otherwise modified sensor outputs. These sensor commands become part of the reference command to a closed loop system which does its best to follow these commands. Since the sensor provides a command to a closed loop actuator system typical closed loop system behavior applies to the actuator portion of the system. Therefore the actuator portion can be analyzed, and a controller designed, using traditional controls techniques.
1.2 Open Loop Aspect
The overall system is open loop because the sensor providing the reference command to the closed loop actuator is not an error signal generated from a feedback sensor but a command based on measurements.
To illustrate this consider a hovering helicopter with gyroscopes capable of measuring yaw, pitch, and roll. In order to continue hovering all rates need to be approximately 0. The gyros measure the rates in 3 axes, the rates are integrated to become positions, and the position are passed as a command to some form of autopilot. The autopilot takes the command and through the use of feedback sensors for rotor speed, tilt, etc. follows the gyro produced command.
The gyro output contains noise, which is then integrated before becoming a command to the autopilot. The motion may be 3 deg/s but the gyro report 3.1 and the autopilot takes out 3.1 so now the helicopter isn't quite where it should be. But the helicopter's hover routine has only the gyros for information regarding the stability of its hover. The integration of the gyro noise over time can lead to large errors which would likely manifest themselves as translational wander. Wander the gyros cannot sense.
For short periods of time feedforward can improve system performance. However, noise in the system will degrade the system performance eventually leading to instability.
Noise characteristics can often be improved through the use of mathematical models, filtering, Kalman filters, or even possibly Neural Networks. Noise can be mitigated but cannot be eliminated entirely.