Feedforward control of robotic arms with noisy actuators

The variability of feedforward controlled robotic arms is influenced by motor noise. This motor noise originates from multiple sources including the motors, transmission and bearings. It is modeled as additive and multiplicative noise. Human arms also suffer from motor noise in feedforward controlled movements, but humans are believed to perform movements that minimize the end-point variability due to their multiplicative input noise. Minimizing the end-point variability for human arms leads to smooth and energy-efficient movements. For robotic arms performing a pick-and-place task the end-point variability should also be as small as possible. The goal of this research is to minimize the end-point variability of feedforward controlled movements in robotic arms. The variability of a movement is determined in simulation using a technique based on the Extended Kalman Filter. The minimum-variability movements are found by performing an optimization of the end-point variability. In simulation it was shown that the minimumvariability movements can make use of the friction in the system, low input torques, and the state of the system. It depends on the exact noise supplied to the system, which aspects are used to minimize the end-point variability. These movements are different from energy-optimized or time-optimized movements. The minimum-variability movements are also applied to an experimental robotic arm set-up. It was found that the variability in the set-up is input-dependent. The minimum-variability movements had lower end-point variability than the energy-optimized or time-optimized movements. This research shows that minimizing the end-point variability for a robotic arm in simulation, leads to movements that have low end-point variability in an experimental set-up too.BMDBioMechanical EngineeringMechanical, Maritime and Materials Engineerin