Many robotics duties, reminiscent of path planning or trajectory optimization, are formulated as optimum management issues (OCPs). The important thing to acquiring excessive efficiency lies within the design of the OCP’s goal perform. In observe, the target perform consists of a set of particular person elements that should be fastidiously modeled and traded off such that the OCP has the specified resolution. It’s usually difficult to stability a number of elements to realize the specified resolution and to grasp, when the answer is undesired, the affect of particular person value elements. On this paper, we current a framework addressing these challenges primarily based on the idea of directional corrections. Particularly, given the answer to an OCP that’s deemed undesirable, and entry to an skilled offering the course of change that may improve the desirability of the answer, our technique analyzes the person value elements for his or her “consistency” with the offered directional correction. This info can be utilized to enhance the OCP formulation, e.g., by rising the load of constant value elements, or decreasing the load of – and even redesigning – inconsistent value elements. We additionally present that our framework can robotically tune parameters of the OCP to realize consistency with a set of corrections.
