Task-based decision making and control of robotic manipulators

Redundancy affords an opportunity to deploy robotic manipulators in a broader variety of tasks owing to extra system resources. Decision making is a procedure that manages these extra resources and utilizes them in a meaningful fashion. Decision making is inherently complex and this complexity is a major hindrance to wider acceptance of redundant manipulators in various robotic applications. This report aims directly at decreasing this complexity by incorporating the concept of task requirements as an integral part of the decision making process. Most traditional Redundancy Resolution Techniques (RRTs) place an undue burden on the user by forcing him to choose a set of performance criteria and assign relative importance to each criterion for a given task. The Task-Based Redundancy Resolution (TBRR) approach presented here changes all that by directly utilizing robotic task requirements in terms of speed, force, and accuracy in the decision making process itself. TBRR searches the null space for configurations that comply with system constraints such as joint travel limits, singularities, and obstacles. TBRR then determines configurations that satisfy task requirements by estimating in real-time robot capabilities using a newly-developed technique called the Vector Expansion method. Finally, TBRR selects, among these configurations, the best solution according to efficiency or any other desired criterion. As a result, TBRR does not require a confusing chore of criteria fusion and thus is easier to use than traditional RRTs. Demonstrations on three geometrically different (6-, 7-, and 10-DOF) spatial robots show that TBRR provides 3% to 147% improvement over traditional RRTs as far as satisfying task requirements. A preliminary effort at integrating a learning method to help determine proper values of subjective parameters significantly reduces the human trial and error effort while showing a 3% improvement over the best set of hand-tuned parameters. Another issue that has not been adequately addressed but is critical to redundant robotic manipulation is force control. Contact tasks, which represent a large portion of robotic tasks, cannot be effectively performed without one form of force control or another. This report illustrates the feasibility of integrating existing force control methods with the TBRR approach. The end result of this research is a task-based decision making and control framework that should enhance task performance of robotic manipulators operating with redundancy in a wider range of robotic tasks.