Follow me

Thursday, October 4, 2012

Fuzzy-GA-Based Trajectory Planner for Robot Manipulator Sharing a Common Wokspace


Emmanuel A. Mechan-Cruz, Alan S. Morris
IEEE Transactions on Robotics, VOL. 22, NO.4, August 2006
Summary
                   Artificial Potential Fields (APF) are useful in describing workspace of manipulators for implementation of fuzzy collision-avoidance strategies. APF is capable of collision avoidance and trajectory planning simultaneously, but has a high risk of being stuck in local minima conditions in certain conditions, therefore in the literature, following Khatib studies, many solutions to avoid local minima have been performed. The authors present a novel fuzzy genetic algorithm (GA) approach to make trajectory planning of two manipulators sharing a common workspace. The application of fuzzy logic for information generation to drive a manipulator to a certain goal was first investigated by Althoefer (1996), but the system appear to no be feasible for two manipulators working simultaneously with a one of them in static condition nearby the goal.
·       Model
The fuzzy-GA-based trajectory planner (FuGABTP) is a simple Genetic Algorithm (GA) trajectory planner (GABTP) where individual fuzzy units provide correction to the displacement of each articulation. Manipulators are modeled as obstacles using a APF method, calculated only for the near vicinity (being a local approach).
The outline see initial configurations being input in both manipulators, which in parallel start first the GABTP and then implement the fuzzy corrections, at that point trajectories are updated and goal is verified if check or not, repeating the sequence if goal is not reached.
The GABTP carries an initial estimation of the motion without considering the presence of possible obstacles, in  free workspace the degree of extension of the manipulator is considered in order to minimize unnecessary displacements of the links (refer to Appendix A): f=1/(e(error/R)), where the error is the distance between goal coordinated and actual coordinate, R is the distance between initial coordinates and actual current coordinates.
Fuzzy correction is achieved by individual Mamdami-type fuzzy units that provide collision avoidance. The obstacle direction can be right is the APF increased or left in the other case. μi are the fuzzy sets, used for mapping μp(d): D à [0;1]. The function to represent the fuzzy sets is asymmetrical triangular functions since these are easily computable. The correction is given according to the fuzzy rules: 1) obstacle on the left and small  APF then SN correction, 2) obstacle on the left and med APF then MN correction; 3) obstacle on the left and medbig APF then MBN correction; 4) if obstacle is on the left and APF is big then BN correction; 5) if obstacle is on the left and APF is 0 then zero correction; 6) if obstacle is on the right and APF is small then correction is SP; 7) if the obstacle is on the right and APF is small then correction is MP; 8) if the obstacle is right and APF is medbig then correction is MBP; 9) is obstacle is right and APF is bit then correction is BP; 10) if obstacle is right and APF is 0 then correction is 0; 11) if obstacle is left and error is 0 then correction is 0; 12) if obstacle is right and error is 0 then correction is 0. The fuzzy set for output correction showing the each acronym graph is report at page 616.
Key Concepts
Fuzzy-GA-Based Trajectory Planner, Planning, Workspace sharing
Key Results
The simulation has been carried on with a 3 d.o.f. and a 7 d.o.f. manipulator both in static and dynamic case, showing success in reaching the goal. The FuGABTP took generally more segment movement but an overall less time to reach the goals if compared with simple GABTP.
The approach demonstrates to be robust and stable, from the set of outputs coming from the model it is further own easy to computer trajectory, velocity and acceleration through kinematics. The parallel planning avoids the need of any further planning.

No comments: