Abstract by Brooke Mosby
lLinearization Approximations for Robotics Control
The design of complicated robots and their application to difficult tasks has generated a need for more sophisticated control algorithms. To simplify and reduce computation time the equations used to model robot motion are modelled by linear system approximations. A common linear approximation is the "Fixed-State approximation", which assumes fixed dynamics between time steps. Although Fixed-State is commonly used, it is expensive to run. A relatively new linearization method, called Coupling-Torque, presents promise in the field of robotics controls. Coupling-Torque assumes that forces imposed from one joint on to another, i.e. coupling terms, are constant between time-steps. This assumption produces equations of motion with variable independence, which uniquely allows for the computations to be parallelized. We are currently analyzing the extent to which coupling terms in the dynamics matrix can be assumed constant, before Coupling-Torque's approximations are completely compromised. This has the potential to significantly reduce computational time and provide and algorithms for modeling the motion of more intricate robots with many joints.