This pull request adds the separation plane parameters as optimization variables in Gurobi optimization, thanks to the new Gurobi 9.1 capable of solving non-convex quadratic constraints. It is tested in Ubuntu 18.04 ROS Melodic, with Gurobi 9.1.2.
I have also added a parameter use_linear_collision_constraint to disable/enable using the plane variables in the optimization. When set to true (which is by default), the algorithm works exactly as before.
Performance: using nonconvex quadratic constraints in Gurobi shows significant increase in computation time (~80ms as compared to ~15ms in the single agent random forest test) and a drop in success rate (<80% in the single agent random forest test).
I have also added some printout (commented by default) which can help to evaluate the performance of the optimization.
This pull request adds the separation plane parameters as optimization variables in Gurobi optimization, thanks to the new Gurobi 9.1 capable of solving non-convex quadratic constraints. It is tested in Ubuntu 18.04 ROS Melodic, with Gurobi 9.1.2.
I have also added a parameter
use_linear_collision_constraintto disable/enable using the plane variables in the optimization. When set to true (which is by default), the algorithm works exactly as before.Performance: using nonconvex quadratic constraints in Gurobi shows significant increase in computation time (~80ms as compared to ~15ms in the single agent random forest test) and a drop in success rate (<80% in the single agent random forest test).
I have also added some printout (commented by default) which can help to evaluate the performance of the optimization.
Hope this is useful.