Closed StoneT2000 closed 2 months ago
The tests failed because of some pathing issues to the new ur5 robot I added for a test - could you rebase this branch onto the latest master?
The goals are generated by sampling joint values within the joint limits without considering obstacles, so the floor is actually cosmetic. Adding constraints to IK would be interesting. The naive way is probably to accept soft constraints as a differentiable external cost function on the joint values q. Another way would be to do gradient projection using the null space of the Jacobian to satisfy the constraints.
Tests for #42
Visualized looks like this:
Not sure why some generated IK goals are under the floor, maybe the robot URDF has poorly set joint limits?
Otherwise seems to all solve anyway: