Robot Kinematics Dynamics

Robot kinematics and dynamics is the part of robotics that describes how a machine's body moves and why.

12 frameworks, from 1850 to today · Robotics · Engineering

still a working view ended · each leads to its entry below

Overview

Robot kinematics and dynamics is the part of robotics that describes how a machine's body moves and why1. Kinematics deals with geometry and motion alone: where the end of an arm is when its joints sit at given angles, how fast it moves, and what joint motions produce a desired motion of the hand1. Dynamics adds force and mass: what torques the motors must supply to make that motion happen, and how the arm responds when it pushes against something. Everything a robot does physically—welding a car body, picking a part off a conveyor, walking across uneven ground—rests on these two bodies of theory. Their history is the story of how engineers learned to write down a machine's geometry and physics in forms a computer can use, and then to compute them fast enough to control a real arm in real time.

Linkages, geometry, and the mechanical heritage

The kinematic study of mechanisms is older than robotics. Nineteenth-century machine theory—associated with figures such as Franz Reuleaux in Germany—catalogued linkages and their motions as a branch of mechanical engineering, analysing how bars, sliders, and gears constrain one another's movement. The vocabulary of this tradition, including the idea of a kinematic chain and the classification of joints by the degrees of freedom they allow, passed directly into robotics when programmable arms appeared. A robot arm is, in this sense, a linkage whose joints are driven by motors and whose configuration is computed rather than set by hand.

The industrial robot of the early 1960s—Unimation's Unimate, installed at General Motors in 1961—was a serial chain of revolute joints, and its programming was essentially teach-and-playback: a human moved the arm through the motions and the machine repeated them. Making such machines do anything more flexible required a mathematical description of the arm's geometry, and that is where the modern subfield begins.

Representing pose: homogeneous transformations and the Jacobian

The first foundational tool was a compact way to describe where a rigid body sits in space. A homogeneous transformation is a 4×4 matrix that combines a rotation and a translation into a single multiplication, so that a chain of links can be described by multiplying matrices together. The convention was formalised for robotics in the 1950s and 1960s, notably in the work of Jacques Denavit and Richard Hartenberg, whose 1955 paper introduced the parameter convention—four numbers per joint—that is still called the Denavit–Hartenberg parameters. With it, the position and orientation of a robot's hand follows from multiplying one matrix per joint.

The companion tool is the Jacobian, a matrix of partial derivatives that relates joint velocities to the velocity of the hand12. It answers the question "if I turn this joint a little, how does the tool move?" and it also exposes singularities—configurations where the arm loses a direction of motion and the Jacobian becomes non-invertible1. The Jacobian became the workhorse of both analysis and control: it converts forces at the hand into torques at the joints, and it underlies the resolved-rate control schemes of the 1970s and 1980s, in which a desired Cartesian velocity is mapped back to joint velocities.

Inverse kinematics: from closed forms to general solvers

Forward kinematics—joint angles to hand pose—is straightforward matrix multiplication13. The hard problem is the inverse kinematics problem: given a desired hand pose, what joint angles achieve it14? For the six-joint arms that dominated industry, closed-form solutions were derived by exploiting geometry5. The classic result is the Pieper solution (from a 1968 Stanford dissertation by D. L. Pieper), which shows that a six-degree-of-freedom arm with three consecutive axes intersecting has a closed-form inverse solution. This structured the design of industrial arms for decades: build the wrist so that its three axes meet at a point, and the inverse problem splits into a position part and an orientation part, each solvable by hand.

Not every arm has such a structure, and closed forms do not exist in general. The alternative is numerical solution—iterative methods that start from a guess and converge on a pose, often using the Jacobian and its inverse or pseudo-inverse. These methods handle arbitrary geometries but can fail near singularities, converge to the wrong branch, or be slow. The tension between closed-form elegance and numerical generality runs through the whole history of the field.

Dynamics: Lagrangian and Newton–Euler formulations

Kinematics says where the arm can go; dynamics says what it costs to get there. Two formulations dominate. The Lagrangian approach, built on energies, derives equations of motion from kinetic and potential energy and yields the standard form in which the joint torques equal inertia times acceleration plus Coriolis and centrifugal terms plus gravity. The Newton–Euler approach, built on forces and moments, works link by link: propagate velocities and accelerations outward from the base, then propagate forces and torques inward from the tool. The two are mathematically equivalent, and the recursive Newton–Euler algorithm, developed in the late 1970s and early 1980s, computes the dynamics in time proportional to the number of joints—a decisive practical advantage over the naive Lagrangian expansion, whose cost grows with the cube of the joint count.

The recursive formulation made real-time dynamic control feasible, and it remains the backbone of modern simulation and control software. The computed-torque control law, which uses the inverse dynamics model to cancel the arm's nonlinearities and reduce the problem to a set of decoupled linear systems, became the canonical demonstration that a good dynamic model buys good control. It is exact only when the model is exact, and its sensitivity to modelling error motivated later adaptive and robust variants.

Screw theory and the product of exponentials

A parallel tradition describes motion using screw theory, in which any rigid motion is a rotation about and translation along a single axis—a screw6. The mathematics goes back to the nineteenth century, to Robert Ball's A Treatise on the Theory of Screws (1876), and was brought into robotics in the 1980s, most influentially by Roger Brockett and then by Richard Murray, Zexiang Li, and S. Shankar Sastry, whose 1994 textbook A Mathematical Introduction to Robotic Manipulation made the product of exponentials formula standard. In this formulation, the forward kinematics of a serial chain is a product of matrix exponentials, one per joint, each built from a screw axis. The approach avoids the bookkeeping of link frames, handles singularities more gracefully, and extends naturally to the Jacobian and to dynamics. It coexists with the Denavit–Hartenberg tradition rather than replacing it; many modern texts teach both.

Parallel mechanisms and closed chains

Serial arms are open chains. Parallel mechanisms close the chain: several legs connect a moving platform to a fixed base, as in the Stewart platform, a six-legged device proposed by D. Stewart in 1965 for flight simulation and later widely used in motion platforms and precision positioning. Closed chains offer high stiffness and load capacity but make kinematics harder, because the constraint equations of all legs must hold simultaneously. The inverse kinematics of a parallel mechanism is often easier than its forward kinematics—the reverse of the serial case—and the field developed its own body of theory for workspace analysis, singularities, and stiffness1. Parallel kinematics became a recognised research area in the 1980s and 1990s, with applications in machine tools and haptic devices.

Floating bases and underactuated systems

Until the 1980s, most theory assumed the robot's base was bolted to the ground. Legged robots and space manipulators break that assumption: their base moves, and the dynamics must account for it. Floating-base dynamics treats the base as an unactuated body coupled to the limbs, which introduces momentum conservation as a constraint and makes control substantially harder. The same mathematics describes underactuated systems—machines with fewer actuators than degrees of freedom—including walking robots, which must exploit passive dynamics and foot contact rather than command every joint. Research on legged locomotion from the 1980s onward, including the Raibert hoppers and later bipeds, drove this area, and it remains central to modern humanoid and space robotics.

The numerical and data-driven turn

The most recent shift is computational. Where closed-form solutions once defined good design, modern practice increasingly relies on numerical optimisation and, more recently, on learned models. Numerical inverse kinematics solvers, trajectory optimisation, and physics engines that integrate the equations of motion at high speed are now standard tools. Machine learning has entered the field as a way to approximate inverse kinematics, dynamics, and contact models from data—useful where analytical models are incomplete, as with flexible objects or complex contact. These methods inherit the classical formulations: a learned dynamics model is still trained against the same equations of motion, and a learned inverse-kinematics network is still judged by the same pose errors.

The field today

Contemporary robot kinematics and dynamics is a layered discipline. The homogeneous transformation and the Jacobian remain the universal language; Denavit–Hartenberg parameters and screw-based products of exponentials are both taught and used; recursive Newton–Euler algorithms underpin real-time control and simulation; and parallel, floating-base, and underactuated formulations extend the classical results to new machine types. The open questions are practical and hard: how to model contact and friction reliably, how to compute dynamics fast enough for high-degree-of-freedom and soft machines, how to guarantee that numerical solvers find the right solution, and how far learned models can substitute for analytical ones without losing the guarantees that made the classical theory useful in the first place.

References

  1. Robot kinematics (introduction, Kinematic equations, Robot Jacobian, Velocity kinematics, Inverse kinematics), Wikipedia
  2. Jacobian matrix and determinant, Wikipedia
  3. Forward kinematics, Wikipedia
  4. Inverse kinematics, Wikipedia
  5. 321 kinematic structure, Wikipedia
  6. Screw theory, Wikipedia

The frameworks of Robot Kinematics Dynamics

Each framework is a way of working the field has used: when it held, and the question it answers.

  1. 1850-Present

    Kinematic geometry of mechanisms

    How do linked bodies constrain each other's motion?

    Fly to its tower in Nousfarer

  2. 1955-Present

    Denavit–Hartenberg parameters

    How can a chain of joints be described by a few numbers?

    Fly to its tower in Nousfarer

  3. 1960-Present

    Lagrangian mechanics

    What torques does a moving mechanism require?

    Fly to its tower in Nousfarer

  4. 1968-Present

    Closed-form inverse kinematics

    What joint angles place the hand at a target?

    Fly to its tower in Nousfarer

  5. 1970-Present
  6. 1970-Present

    Numerical inverse kinematics

    How can a target pose be found by iteration?

    Fly to its tower in Nousfarer

  7. 1975-Present

    Recursive Newton–Euler dynamics

    How can required torques be computed quickly?

    Fly to its tower in Nousfarer

  8. 1980-Present

    Computed-torque control

    How can a model make a machine follow a path?

    Fly to its tower in Nousfarer

  9. 1980-Present

    Screw theory

    How can any rigid motion be seen as one twist?

    Fly to its tower in Nousfarer

  10. 1985-Present

    Parallel manipulator

    How do several legs jointly position a platform?

    Fly to its tower in Nousfarer

  11. 1985-Present

    Underactuation

    How can a machine move with too few motors?

    Fly to its tower in Nousfarer

  12. 2000-Present

How the ideas connect

Each arrow runs from an earlier idea to the later one it shaped. The list below says what passed between them.

Kinematic geometry of mechanismsDenavit–Hartenberg parametersLagrangian mechanicsClosed-form inverse kinematicsHomogeneous transformation and Jacobi…Numerical inverse kinematicsRecursive Newton–Euler dynamicsComputed-torque controlScrew theoryParallel manipulatorUnderactuationComputational and data-driven kinemat…

Beyond the field

Where Robot Kinematics Dynamics's ideas came from in other fields, and what they led to there: the strongest 12 of 74 links.

A city in Nousfarer: a red-carpeted stair climbing terraces to a crowned tower on an acropolis, more towers round it, forest and snowy mountains beyond.
A city in Nousfarer, the game played in the browser.

Robot Kinematics Dynamics is a planet in Nousfarer

Its city has a tower for each of these 12 frameworks. Land, walk in, and climb one concept at a time.

Land on Robot Kinematics Dynamics

Related fields

Fields its ideas are linked to

More in Robotics

See where Robot Kinematics Dynamics sits on the Map of Human Knowledge