Work in the ECHORD GOP project continued with a close interaction of the project partners in France and Germany, based on regular (at least bi-weekly) Skype meetings of all project members and some personal meetings of the responsible persons in the project.  

A central topic of discussion were the pros and cons (in the context of the problem class addressed in GOP) of feasible optimization techniques with respect to optimization techniques that allow infeasibility in the course of the solution. The direct multiple shooting optimal control code MUSCOD of IWR, Heidelberg University, uses a specially tailored SQP (sequential quadratic programming) method for the solution of the unserlying nonlinear programming problem (NLP) which is only guaranteed to satisfy all constraints in the solution but can still violate constraints in the iterations before which is considered to be advantageous in the case of complex constraint sets. However, this requires the constraints to be formulated in such a way that they can also give back a meaningful value in the case where they are violated. But this is not the case in the efficient constraint evaluation routines such as KCD (Kineo Collision detector) used at LAAS which give only a qualitative but no quantitative answer in the case of constraint violation. So the coupling of optimal control and motion planning techniques is not possible in a straightforward way. Two approaches are discussed:

1.     adding a computation (or at least an estimation) of the constraint violation in the motions planning functions evaluating distance to obstacles

2.     using feasible optimization solvers such as interior point techniques for optimization instead. 

Work at LAAS during this period has concentrated on installing and evaluating an own optimization library called Roboptim based on feasible optimization techniques. Trajectories are discretized using splines. For the resulting NLP, the optimization solvers IPOPT and CFSQP are implemented.  The setup of the optimal control problem of the humanoid robot is in progress, as well as the direct coupling with KCD which is easily feasible iin this case.

Work at IWR, Heidelberg in the ECHORD project has focused on the practical side, i.e. the set up of the robotic demonstrator and some modeling aspects since the position responsible for the optimal control aspects has been vacant for this two-month period.  Optimal control work will be taken up again on October 1^st.

The KUKA robot arm has arrived at the beginning of August and has been set up and has been thoroughly tested.  We have also mounted the pneumatic gripper along wit its pressure control system.  The robot has been set up on a specially manufactured solid table in order to be able to run the robot at full speed. We have also started to familiarize with the theory of realtime robot control via XML strings on ethernet. Example trajectories have been executed on the robot. Different options for the dynamic test environment for the KUKA arm have been evaluated. In addition we have worked on expanding the VRML model of the KUKA robot arm to include the gripper and integrated the VRML model and the kinematik chain into the Kineo software. The integration of the Kineo library into own C++ code has been started.