Real-time generation of fast, torque-efficient motions

We present a method for generating fast, torque-efficient motions in real-time. Given a dynamic model of a robot, we first obtain a reasonably large set of minimum torque motions for arbitrary starting and ending configurations, that also achieve minimum time without actuator saturation; these motions are obtained through a Lie group dynamics-based optimization procedure previously developed by the authors. After classifying the minimum torque motions according to the region of the workspace traversed, a principal component analysis is performed for each class of motions. The dominant principal components are then used as basis functions in a simple interpolating scheme for generating torque-efficient motions between arbitrary starting and ending configurations. Results obtained for a typical industrial manipulator clearly demonstrate the performance advantages over traditional spline interpolation methods