Optimization based soft-sensing framework for modularized robots

System state information for feedback control and pose planning in robotic systems often requires multiple torque sensors that can increase the overall system complexity and cost. Using available module internal sensor data and rigid body simulation based observers, we can minimize the number of hardware sensors required to estimate the actual robot state. In the case that the state unknowns significantly impact the system dynamics, traditional joint torque observer methods may degrade due to un-modeled nonlinearities. We propose a real-time soft sensing framework using optimization techniques and comprehensive simulation models to estimate a wide variety of possible system unknowns using minimal hardware sensors.