Forward kinematics

Forward kinematics (FK) is the calculation of a robot's end-effector position and orientation from its joint positions, meaning the angles of rotating joints and the extensions of sliding ones. It chains one rigid transform per joint, from the robot's base to its tool, and for a serial arm it always has exactly one answer. It is the reverse of inverse kinematics, which solves for the joint positions that reach a target pose.

What is forward kinematics?

A robot arm's controller knows the angle of every joint. Forward kinematics answers the next question: given those angles, where is the end-effector, and which way is it pointing? It is the reverse of inverse kinematics, which starts from a target pose and solves for the joint angles.
For a serial arm, forward kinematics always has exactly one answer, and it is fast to compute. Inverse kinematics can have many solutions or none, which is why it is the harder problem.

Key takeaways

  • Forward kinematics maps joint angles to the end-effector's position and orientation.
  • It chains one rigid transform per joint, from the robot's base to its tool.
  • It turns logged joint states into end-effector paths you can check against camera views.

How it works

Each link of the arm is a rigid body, and each joint adds one rotation or translation. Forward kinematics multiplies the transforms for every joint, from the base to the end-effector, to get the end-effector's 6D pose. The geometry comes from the robot's description, written as Denavit-Hartenberg parameters or, in ROS, a URDF file. In ROS 2, the robot_state_publisher node does this continuously: it reads joint states and the URDF and publishes the pose of every link as tf transforms.
Forward kinematics chains one transform per joint to get the gripper's pose.

See how far and how fast the gripper traveled in every episode

Forward kinematics turns joint data into motion you can see. Watch for:
  • end-effector paths that jump between frames
  • peak end-effector speeds far above the rest of the dataset
  • episodes where the arm travels far more, or less, than the task needs
In Voxel51, DROID holds 1,000 real-world manipulation episodes with motion stats such as peak end-effector speed and total end-effector displacement.

Robotics Droid

1,000 episodes

Real-world robot manipulation episodes from DROID, stored as MCAP with three synced cameras and fields for task prompt, success, and number of grasps.

Try this: Sort episodes by total_ee_displacement_m and open the highest and lowest. Check whether the arm's travel on camera fits the ranking.

Explore DROID in Voxel51

Free with a work or .edu email.

Go deeper

Why it matters

Robot logs usually record joint positions, and many datasets also record the end-effector pose. Forward kinematics connects the two. It turns joint data into a path through space that can be plotted, measured, and overlaid on camera frames. It also gives a consistency check: if the computed end-effector path disagrees with the recorded one, or with where the gripper appears on camera, something is off in the calibration, the robot model, or the timestamps.
Joint positions go in, and the end-effector pose comes out.

Frequently asked questions

What is the difference between forward and inverse kinematics?

Forward kinematics goes from joint angles to the end-effector pose and has one answer. Inverse kinematics goes from a desired pose to joint angles and can have several answers or none.

What are Denavit-Hartenberg parameters?

They are a compact convention that describes each link of a robot arm with four numbers, which together define the transform from one joint to the next.

Does forward kinematics account for forces?

No. Kinematics covers positions and motion only. Forces and torques belong to dynamics.
black and white photo of Jesse Mostipak
Director of Growth
Jesse Mostipak is the Director of Growth at Voxel51, where the work is helping humans find and trust what the brand knows, and teaching the Google knowledge graph and the LLMs answering on their behalf to do the same. That question, how knowledge gets built inside a system, is one Jesse has been chasing for years. Earlier versions of it ran through a New York City high school science classroom, data science and machine learning, and developer relations at Kaggle, Posit (formerly RStudio), and Baseten. The answer doesn't change much depending on whether the learner is a teenager, a software engineer, or a knowledge graph. Jesse holds a Master's in Education from CUNY Hunter College. LinkedIn
See all articles by Jesse Mostipak
Last updated November 19, 2026

Building visual or physical AI?

Let's talk.