Forward Kinematics using Orocos KDL

Sarvagya Vaish
5 min readMar 10, 2017

This post briefly describes how to define coordinate systems for robots and then explains how to use Orocos KDL to perform forward kinematics.

A Simple Robot

To make things a little easier, we’ll be using the simple robot model shown below. It has four links (L1 — L4) and three degrees of freedom. Link 1 is fixed to the ground while Link 2 and Link 3 can rotate in the screen plane. Link 4 can rotate around like a periscope, about the axis shown in the schematic.

Our goal is to find the pose of the last link, the end effector, given the measurement of three joint angles (J1, J2, J3).

Links and Joints

Defining Coordinate Frames

Let’s start with defining the coordinate frames for each link. F0 is our reference coordinate frame, the frame relative to which we would like to find our end effector’s pose, F4.

I am using the widely used Denavit-Hartenberg convention to define the intermediate links. Specifically, it says that the Z axis should be in the direction of the joint axis.

Coordinate frames attached to links
  • F1 is attached to Link 1 with the Z axis aligned to Joint 1, coming out of the plane.
  • F2 is attached to Link 2 with the Z axis aligned to Joint 2, coming out of the plane.
  • F3 is attached to Link 3 with the Z axis aligned to Joint 3, along the axis of rotation for Link 4.
Robot pose for joint angles -45°, 90°, 180°

Here’s another image to illustrate how the coordinate frames are attached to the links and joints.

  • Joint 1 is rotated -45°
  • Joint 2 is rotated 90°
  • Joint 3 is rotated 180°

Assuming that the origin is at F0, the end effector’s position, F4, is roughly at (-1.5, 3.8, 0). We’ll calculate this as well the the full pose using forward kinematics in the next section.

Forward Kinematics

What is Orocos KDL?

Orocos (Open Robot Control Software) is a suite of libraries for robot arm control. Orocos KDL (Kinematics and Dynamics Library) provides the ability to create kinematic chains to perform forward and inverse kinematics.

Step 1: Creating Segments

In KDL, kinematic chains consist of segments. Segments are essentially the links of the robot. A segment is a combination of a Joint and a Frame. The joint tells the segment how a frame moves as the joint angle changes. For example, it specifies if the frame rotates about the Z axis.

Frames are specified relative to the preceding frame:

  • F1 is defined relative to F0
  • F2 is defined relative to F1
  • F3 is defined relative to F2
  • F4 is defined relative to F3

Let’s create an empty chain to which we’ll add the segments.

Chain kdlChain = Chain();

For the first segment, note how F1 is oriented the same way as F0, just located at (0, 1, 0). Hence, we can initialize our frame using a simple vector. Also note that F1 doesn’t move relative to F0. In this case we’ll use a Joint::None while creating the segment.

Joint joint1(Joint::None);
Frame frame1 = Frame(Vector(0.0, 1.0, 0.0));
kdlChain.addSegment(Segment(joint1, frame1));

Similarly, for F2 there is again no change in orientation of the frame, just a translation to (0, 2, 0). Link 2 however rotates about the Z axis of the previous frame, so we’ll use Joint::RotZ.

Joint joint2(Joint::RotZ);
Frame frame2 = Frame(Vector(0.0, 2.0, 0.0));
kdlChain.addSegment(Segment(joint2, frame2));

F3 is rotated about the X axis by -90° and then translated 2 units about the new Z axis. We can multiply two intermediate frames to construct the final one.

Joint joint3(Joint::RotZ);
Frame frame3 = Frame(Rotation::EulerZYX(0.0, 0.0, -M_PI / 2)) *
Frame(Vector(0.0, 0.0, 2.0));
kdlChain.addSegment(Segment(joint3, frame3));

F4 first reverses the previous rotation and is then translated 1 unit in the X and Y directions.

Joint joint4(Joint::RotZ);
Frame frame4 = Frame(Rotation::EulerZYX(0.0, 0.0, M_PI / 2)) *
Frame(Vector(1.0, 1.0, 0.0));
kdlChain.addSegment(Segment(joint4, frame4));

Step 2: Joint Angles

Next we’ll construct a joint angles variable that will contain the three angles for which we wish to perform forward kinematics.

Robot pose for joint angles -45°, 90°, 180°
JntArray jointAngles = JntArray(3);
jointAngles(0) = -M_PI / 4.; // Joint 1
jointAngles(1) = M_PI / 2.; // Joint 2
jointAngles(2) = M_PI; // Joint 3

Step 3: Forward Kinematics

Finally, we can run forward kinematics for the joint angles in step 2, using the chain we constructed in step 1. This will give is the end effector pose.

ChainFkSolverPos_recursive FKSolver =
ChainFkSolverPos_recursive(kdlChain);
Frame eeFrame;
FKSolver.JntToCart(jointAngles, eeFrame);

Forward kinematics solution (contents of eeFrame):

-0.7071   -0.7071    0.    -1.414
-0.7071 0.7071 0. 3.828
0. 0. -1. 0.
0. 0. 0. 1.

Let’s verify a couple of things against the diagram.

  • the end effector’s position is roughly at (-1.5, 3.8, 0); the calculated position is (-1.414, 3.828, 0.)
  • the end effector frame’s Z axis unit vector is pointing along F0’s -Z Axis
  • the X axis unit vector is (-0.7071, -0.7071, 0.) which checks out too since in the image it’s in the screen plane and pointing down and to the left — negative x, negative y relative to F0
  • similarly, the Y axis unit vector is (-0.7071, 0.7071, 0.) which means it should be in the screen plane and pointing up and to the left — negative x, positive y relative to F0

The code

Code is available on Github.

Here’s the relevant code all at once:

Acknowledgements

for technical proofreading.

--

--