There is an option to avoid joint limits in simIK.setGroupFlags:
simIK.group_avoidlimits: joint limits are actively avoided
A more customizable approach to drive the robot configuration towards "better" poses could be of introducing a secondary task, e.g. to attract the elbow up, without affecting the accuracy of the primary task.
Another more advanced approach known in robotics literature consists in -provided the robot has redundant degrees of freedom for the task- computing the
null space of the
jacobian matrix. The
span of the null space will give you all the joint directions in which to move while keeping the tip-target error to zero, e.g. the so called "self motion" can be obtained by this. The simIK plugin offers various way to obtain the jacobian matrix (simIK.computeJacobian, simIK.computeGroupJacobian, with a callback argument to simIK.handleGroup, ...) then CoppeliaSim provides a linear algebra library (
matrix.lua) to conveniently work with vectors and matrices.
