#robotics Logs

Oct 24 2015

18:38 veverak please
18:38 veverak somebody tell me
18:38 veverak that this is bullshit product and I don't want it
18:49 wolfmanjm I can;t live without my 4 channel Sallae one :)
19:16 SpeedEvil get a FPGA dev-kit, and learn to DIY
19:50 superlou Are general robotics questions ok?
19:51 Snert__ yes ask away
19:52 superlou Snert_, thanks. Looking for some direction: I'm trying to solve a system of parallel robot (started in http://imgur.com/NE0fpnQ) where there are three servos with an arm of lenght l, each servo radially spaced 120 degrees apart. Each servo arm is connectd via a 1 DOF rotational joint to another arm of lenght m, ending in a spherical joint C. Each of the final joints are fixed together, forming an equilateral triangle with sides len
19:52 superlou gth n. By adjusting
19:53 superlou the servo positions, you can drive the triangle with 3 degrees of freedom, primarily up/down, tilt, and roll. However, trying to solve for the locations of C is turning into a nightmare. No matter what i've tried I end up with a huge mess of terms. I've been trying to think of some way to take advantage of the symmetry without much luck
19:53 superlou (stupid enter key)
19:53 superlou sorry. I thought the direct kinematics would be easier to solve, but I'm banging my head trying to find a more elegant solution, then trying to solve 3 systems of equations with something like 30 terms each
19:54 Tom_itx rue did some things like that i believe
19:54 superlou I can see there are three unknowns (the angle at the second arm with respsect to ground), and three equations
19:55 superlou Tom_itx, i've seen a bunch of resources on parallel delta robots like http://www.ohio.edu/people/williar4/html/pdf/DeltaKin.pdf
19:55 superlou but I'm feeling dense and can't figure out how to apply it to this case
19:55 superlou Pretty excited about my robot: https://vimeo.com/136165642. But he can't get any smarter until I have some sort of IK solution.
19:59 Snert__ that's exactly like a swashplate on an rc helicopter.
19:59 Snert__ pretty kool
20:00 superlou Snert_, I think most swash plates only use two axis?
20:00 Snert__ some do...but the majority is 3.
20:00 superlou huh
20:01 Snert__ there's a split some like 2 main and a lever servo thing. I prefer 3 in delta.
20:01 superlou Snert_, oh, those are going to be ball joints, with the central pillar making it go straight up and down
20:01 superlou In this, the center of the delta actually translates a bit since the three arms can only move in a plane
20:02 superlou Which is making the IK solution lame since I can't just pick an arbitrary delta position and solve for it, first have to figure out if it's viable
20:02 superlou So I think I may try to do the IK solution first rather than DK. I want to control the tilt, roll, and Z position of the delta, so I need to find a solution where that delta has it's points in the three 120 degree space planes
20:03 superlou does that make any sense?
20:03 Snert__ you could throw a 3axis rc heli gyro in the mix for $9.95 and have a level top plate always/
20:04 Snert__ but anyways, can't help. I do themath like that.
20:04 Snert__ dont do themath, that is.
20:04 superlou Snert_, i was considering mounting an accelerometer to the delta and doing some sort of learning algrotihm where it randomly moves the servos, measures angle, and makes a look up table
20:04 superlou unfortunately, won't give Z data
20:32 SpeedEvil superlou: gyroscopes are cheap
20:32 superlou Hey SpeedEvil! Have we met from ##electronics?
20:32 SpeedEvil And don't drift very fast.
20:33 SpeedEvil I don't know
20:33 superlou Gyro doesn't really solve the problem, unforutnately
20:33 superlou and i'd like open loop control
20:33 SpeedEvil It's lots better than accellerometer
20:33 SpeedEvil Though you need both to get non-drifty in many ways
20:33 superlou probably
20:33 superlou yeah
20:34 SpeedEvil But fortunately, cheap
20:34 superlou But my problem is still math. I shouldn't need either
20:34 SpeedEvil ah
20:34 superlou Was hoping to make this an open loop design at this point
20:35 superlou I"m sitting here staring at a piece of paper trying to find a way to define the tilt and roll of a triangle in space
20:35 superlou It's starting to hurt
20:35 superlou htought i was making progress using a point (centroid of traingle) and the normal vector
20:35 superlou but then you need ot define the orientation of the trianl with another vector
20:37 SpeedEvil 'yes, it's going to cost twice as much to do it this way but I understand the maths'
20:38 Jak_o_Shadows What are you trying to do?
20:38 superlou Jak_o_Shadows, inverse kinematics for this robot: https://vimeo.com/136165642
20:39 superlou I spent today trying to do direct kinematics analytically so i could then use a numeric technique to come up with IKs. No luck. The equations are super gross.
20:39 superlou I was able to implement the DKs numerically, so the theory is somewhat ok.
20:39 superlou Now I'm trying to directly come up with the IKs by first orienting the robot's "head" in an achievable position/oreintation for the actuators
20:40 SpeedEvil Neural networks!
20:40 SpeedEvil that way you can come up with a solution that no human knows how it works.
20:40 Jak_o_Shadows So you've got half a stewart platform
20:40 superlou SpeedEvil, https://github.com/superlou/deskbot/tree/master/kinematics (I have been trying Neural nets)
20:40 SpeedEvil http://rocknrollnerd.github.io/ml/2015/05/27/leopard-sofa.html
20:40 superlou unfortunately, when i test it they don't work out
20:41 superlou Jak_o_Shadows, essentially, though the actuation axis are somewhat different
20:41 superlou And with a stewart, you have 6-dof, in this I have 3, and many of them are coupled, making it hard to work backwards
20:44 Jak_o_Shadows Your 3 DOF are height, pitch, roll?
20:46 superlou Jak_o_Shadows, yep. The yaw and x/y translation are uncontrolled but effected
20:47 superlou however, my definitions of pitch and roll aren't great, because the undesired yaw will cause them to change I think
20:47 Jak_o_Shadows Because you're using imperfect linkages to do linear motion?
20:47 superlou Even assuming perfect linkages
20:48 Jak_o_Shadows If you assume perfect straight-up-down motion of the things, you only have those 3 DOF
20:48 superlou if they were planar actuators, then it would be simpler to solve, but you'd still need the ball joint at the head i think
20:49 Jak_o_Shadows Yeah.
20:49 superlou It's turned out to be much tricker than i thought it would be
20:50 superlou I think I can't think of the DOFs as pitch and roll, but really a single normal vector
20:50 superlou from a point, where i'm controlling the point's Z position
20:50 superlou that point is the centroid of the triangle (the head)
21:37 rue_shop3 who is guy
21:45 Jak_o_Shadows ?
22:06 rue_house guy needs help
22:07 Tom_itx you're too late
22:07 Tom_itx he left
22:24 rue_house usualy story