Robot Kinematics and Motion Representation

Introduction

Robot kinematics studies the geometry of motion without considering the forces and torques that cause it. It provides the mathematical relationship between joint variables and the position and orientation of a robot’s end-effector. In industrial robots, this relationship is essential for programming, trajectory planning, collision avoidance, calibration, control, and workspace analysis.

A robot mechanism is commonly represented as a serial chain of rigid links connected by revolute or prismatic joints. The configuration of the robot is described by its joint variables. For a revolute joint, the variable is an angle, usually denoted θ; for a prismatic joint, it is a displacement, usually denoted d. Kinematic analysis has two central problems:

• Forward kinematics: determining end-effector pose from known joint variables.
• Inverse kinematics: determining joint variables required to achieve a specified end-effector pose.

Velocity analysis introduces the Jacobian matrix, which maps joint velocities to linear and angular velocities of the end-effector. The Jacobian also reveals singularities, force transmission properties, and local manipulability.

Learning Objectives

By the end of this chapter, the reader should be able to:

1. Distinguish position, orientation, pose, configuration, and workspace.
2. Represent rigid-body motions using rotation matrices and homogeneous transformations.
3. Construct a robot model using the Denavit–Hartenberg convention.
4. Derive and interpret forward kinematic equations.
5. Formulate inverse kinematic solutions and recognize multiple or nonexistent solutions.
6. Derive the Jacobian and use it to relate joint and Cartesian velocities.
7. Identify singular configurations and explain their practical consequences.
8. Analyse workspace and manipulability for robot design and operation.

1. Fundamental Kinematic Concepts

1.1 Configuration and pose

The configuration of a robot is the complete set of joint variables. For an n-joint robot, it is represented by

where qᵢ is either a joint angle in radians or a joint displacement in metres. The configuration space is the set of all allowable q values, including mechanical limits and, where relevant, self-collision restrictions.

The pose of a rigid body consists of its position and orientation relative to a reference frame. Position is a three-dimensional vector p = [x y z]ᵀ. Orientation describes how the body frame is rotated relative to the reference frame.

A pose is therefore written as

where φ is any three-parameter orientation representation, such as roll–pitch–yaw angles. This notation is useful but depends on the chosen orientation coordinates. A homogeneous transformation is generally more robust for composing motions.

1.2 Reference frames

A coordinate frame consists of an origin and three mutually perpendicular unit axes. A frame attached to the robot base is commonly denoted {0}, and a frame attached to the end-effector is denoted {E}. The notation ⁰T_E means the pose of frame {E} expressed relative to frame {0}.

A frame assignment must be consistent. Its axes should be right-handed, and all transformations must use the same length and angular units. In practical robot programming, positions are often expressed in millimetres while mathematical models use metres; mixing these units produces serious errors.

2. Rotation and Rigid-Body Transformation

2.1 Rotation matrices

A rotation matrix R is a 3 × 3 matrix whose columns are the unit axes of the rotated frame expressed in the reference frame. It satisfies

where I is the identity matrix. The transpose is therefore the inverse: R⁻¹ = Rᵀ.

The elementary rotations are

Angles α, β, and γ are measured in radians in calculations. Rotation multiplication is not commutative: in general, R₁R₂ ≠ R₂R₁. Consequently, the order in which rotations are applied must be stated.

2.2 Homogeneous transformations

A homogeneous transformation combines rotation and translation:

where R is the relative orientation and p is the relative position. The transformation acts on a homogeneous point [x y z 1]ᵀ. For frames {A} and {B}, the matrix ᴬT_B converts coordinates from frame {B} to frame {A}:

The inverse transformation is

This formula follows from R⁻¹ = Rᵀ and is valid for rigid-body transformations.

Transformations compose by matrix multiplication. For a serial robot,

Each matrix must describe the motion from one adjacent frame to the next. The product order is essential.

2.3 Orientation representations

Rotation matrices avoid singularities but contain nine numbers subject to six constraints. Other representations include:

• Roll–pitch–yaw angles: intuitive, compact, but sequence-dependent and vulnerable to gimbal lock.
• Axis–angle representation: specifies a unit axis and rotation angle; useful conceptually and computationally.
• Quaternions: compact and numerically effective for interpolation and control, but require unit normalization and have the equivalent-sign property q and −q representing the same orientation.

No representation is universally best. Rotation matrices are particularly suitable for derivations involving frame transformations, while quaternions are often preferred in real-time orientation interpolation.

3. Denavit–Hartenberg Representation

The Denavit–Hartenberg (DH) convention provides a systematic method for assigning frames to serial manipulators. Each link-to-link transformation is described by four parameters:

In the standard DH convention, frame {i} is obtained from frame {i−1} by

which expands to

For a revolute joint, θᵢ is variable and dᵢ is constant. For a prismatic joint, dᵢ is variable and θᵢ is constant. The remaining parameters describe fixed link geometry.

Frame-assignment procedure

1. Number joints and links consistently.
2. Place zᵢ along the axis of joint i.
3. Place xᵢ along the common normal between zᵢ and zᵢ₊₁; if axes intersect, choose xᵢ perpendicular to their plane.
4. Place the origin at the intersection of xᵢ with zᵢ₊₁, or at a logical reference point when axes are parallel.
5. Complete each right-handed frame using the y-axis.
6. Record aᵢ, αᵢ, dᵢ, and θᵢ.
7. Multiply the adjacent transformations in order.

Different sources use standard and modified DH conventions. They are both valid, but their frame assignments and transformation formulas differ. A model must never mix the two conventions.

4. Forward Kinematics

Forward kinematics calculates the end-effector pose from q. For a DH model with n joints,

The upper-left 3 × 3 submatrix gives orientation, and the upper-right 3 × 1 column gives position.

Worked Example: Two-link planar arm

Consider a two-revolute-joint planar arm with link lengths l₁ and l₂. Both joints rotate about parallel z-axes, and motion occurs in the x–y plane. The position of the wrist is

The planar orientation of the second link is

For l₁ = 0.40 m, l₂ = 0.30 m, θ₁ = 30°, and θ₂ = 45°:

x = 0.40 cos30° + 0.30 cos75° ≈ 0.3464 + 0.0776 = 0.4240 m.

y = 0.40 sin30° + 0.30 sin75° ≈ 0.2000 + 0.2898 = 0.4898 m.

Thus the end-effector position is approximately (0.424, 0.490) m and its orientation is 75°. The calculation assumes ideal rigid links, exact joint angles, and no compliance or calibration error.

5. Inverse Kinematics

Inverse kinematics determines q for a desired pose. Unlike forward kinematics, it may have multiple solutions, one solution, or no solution. It is generally nonlinear and can be computationally demanding.

For the planar two-link arm, define r² = x² + y². Applying the cosine rule gives

The joint-angle solutions are

where c₂ is the calculated cosine value. The corresponding θ₁ is

The ± sign produces the elbow-up and elbow-down configurations. A real solution requires −1 ≤ c₂ ≤ 1. If |c₂| > 1, the target lies outside the reachable workspace, subject to joint limits.

For spatial six-axis robots, inverse kinematics may be solved analytically, numerically, or by hybrid methods. Analytic solutions are fast and reveal all branches but are robot-specific. Numerical methods are more general but depend on initial guesses and may converge to an undesirable branch.

A common numerical approach uses iterative correction:

where qₖ is the current estimate, J⁺ is the pseudoinverse of the Jacobian, and eₖ is a pose error represented consistently in position and orientation coordinates. This method is local and may fail near singularities or joint limits.

6. Velocity Kinematics and the Jacobian

The Jacobian is the differential mapping between joint rates and end-effector twist. For a six-dimensional twist,

where v is linear velocity in m/s, ω is angular velocity in rad/s, q̇ is the vector of joint rates, and J is a 6 × n Jacobian. The upper three rows map joint rates to linear velocity; the lower three map them to angular velocity.

For a revolute joint i, let zᵢ₋₁ be its axis expressed in the base frame, oᵢ₋₁ its origin, and o_n the end-effector origin. Its Jacobian columns are

For a prismatic joint,

These expressions assume all vectors are expressed in the same coordinate frame and that the end-effector point is fixed relative to the final link.

Planar Jacobian example

For the two-link planar arm,

and

The first column describes the Cartesian velocity caused by unit motion of joint 1, while the second describes that caused by unit motion of joint 2.

The determinant is

for this square 2 × 2 planar Jacobian. It becomes zero when θ₂ = 0 or π, meaning the two links are collinear. These are singular configurations.

7. Singularities

A singularity is a configuration at which the Jacobian loses rank. At such a configuration, one or more Cartesian velocity directions cannot be produced by finite joint velocities. Conversely, attempting to produce certain Cartesian velocities may require extremely large joint rates.

For a square Jacobian, singularity occurs when det(J) = 0. For a rectangular Jacobian, rank deficiency is assessed using rank, singular values, or the determinant of J Jᵀ where appropriate.

Typical consequences include:

• loss of instantaneous degrees of freedom;
• large joint velocities near the singularity;
• poor numerical conditioning of inverse kinematics;
• reduced force capability in particular directions;
• sudden changes between inverse-kinematic branches.

A robot controller should not simply command an exact singular pose without assessing speed, load, and control consequences. Damped least-squares methods improve numerical behaviour:

where λ > 0 is a damping factor. Increasing λ reduces velocity amplification but introduces tracking error. The damping should be selected according to conditioning, speed requirements, and control accuracy.

8. Workspace

The workspace is the set of end-effector poses or positions achievable within joint limits. A position workspace ignores orientation; a dexterous workspace includes the required orientations.

For an unrestricted two-link planar arm, the radial reach satisfies

where r = √(x² + y²). The region is an annulus, although joint limits may remove sectors from it. Obstacles, tool geometry, cable routing, and collision constraints further reduce the usable workspace.

Workspace analysis is important when selecting a robot for tasks such as welding, palletizing, machining, or inspection. A target lying within nominal reach may still be infeasible because the required orientation is unavailable or because the robot must pass through a singular or collision-prone posture.

9. Manipulability and Conditioning

Manipulability describes how effectively joint motion produces Cartesian motion. Yoshikawa’s velocity manipulability measure is

for a full-row-rank Jacobian. A large value generally indicates that motion can be generated in many directions without extreme joint speeds; w = 0 indicates a singularity. Because linear and angular velocity rows have different units, a combined measure requires appropriate scaling or normalization.

The singular value decomposition is

The singular values in Σ describe directional velocity gains. The ratio

is the condition number when σ_min > 0. A large κ indicates anisotropic and poorly conditioned motion. A condition number of 1 represents isotropic scaling, although such a configuration may not be attainable in a particular robot.

Manipulability is useful for:

• choosing robot postures during task execution;
• placing workpieces and fixtures;
• comparing robot designs;
• avoiding configurations that amplify sensor or model errors;
• planning trajectories with smooth velocity requirements.

10. Applied Case: Cartesian Pick-and-Place Cell

Background

Consider an illustrative six-axis industrial robot required to pick components from a conveyor and place them into a fixture. The fixture is within the robot’s nominal position workspace, but the tool must approach vertically and maintain a specified orientation.

Problem

Initial programming based only on point coordinates produces occasional high joint speeds near the approach pose. Some target orientations also cause the wrist axes to become nearly aligned, producing a wrist singularity.

Analysis

The engineering team first establishes a consistent base frame, tool frame, and work-object frame. A DH model is created and verified against measured link dimensions. Forward kinematics confirms the tool pose for several measured joint configurations. The Jacobian is then evaluated along the proposed trajectory.

The smallest singular value decreases sharply during the vertical approach. This indicates poor conditioning. Although the target position is reachable, the required orientation causes the robot to pass close to a singular posture.

Approach and solution

The team modifies the approach path by adding an intermediate waypoint and rotating the tool slightly during the approach. Inverse kinematics is solved with the previous configuration as the initial estimate, helping maintain the desired solution branch. A damped least-squares inverse is used near low manipulability. Joint limits and collision checks are included in the trajectory planner.

Outcome

The revised trajectory avoids the worst singular region, reduces peak joint velocity, and maintains the required placement accuracy. Cycle time increases slightly, but the process becomes more reliable and reduces emergency stops.

Learning points

1. Position reachability does not guarantee pose reachability.
2. Jacobian analysis should accompany inverse-kinematic planning.
3. Frame definitions and tool calibration are as important as algebraic equations.
4. A small path modification can substantially improve conditioning.
5. Reliability and safety may justify a modest increase in cycle time.

11. Implementation Practices and Limitations

A reliable kinematic implementation should:

• document frame conventions and units;
• validate each transformation independently;
• compare forward-model predictions with measured poses;
• use atan2 rather than a single-argument arctangent when quadrant information matters;
• enforce joint limits and configuration-branch preferences;
• monitor Jacobian singular values during motion;
• distinguish commanded, measured, and calibrated tool frames;
• account for encoder zero offsets and link-geometry errors.

Kinematic models are idealizations. They usually neglect structural flexure, backlash, bearing clearance, thermal expansion, gear compliance, payload-induced deflection, and calibration uncertainty. These effects become important in high-speed machining, precision assembly, and long-reach manipulators.

Safety systems must not rely solely on a mathematical kinematic model. Industrial robots require validated limits, protective guarding or monitored space, emergency-stop systems, and risk assessment under applicable machinery and workplace regulations. A kinematic solution is not by itself a safe motion command.

Contemporary systems increasingly combine geometric models with sensor feedback, digital twins, optimization, and machine learning. Data-driven corrections can improve calibration and compensate for repeatable errors, but they should not replace interpretable geometric verification where safety and traceability are required.

Chapter Summary

Robot kinematics describes motion without forces. A robot configuration is defined by its joint variables, while a pose combines position and orientation. Rotation matrices represent orientation, and homogeneous transformations combine rotation and translation in a form suitable for serial multiplication.

The Denavit–Hartenberg convention assigns four parameters to each link-to-link relationship and provides a systematic forward-kinematic model. Forward kinematics maps joint variables to end-effector pose. Inverse kinematics performs the reverse mapping and may yield multiple, unique, or nonexistent solutions.

The Jacobian maps joint velocities to end-effector linear and angular velocities. Its rank and singular values reveal singularities, conditioning, and manipulability. Singular configurations can cause loss of motion directions, excessive joint rates, and numerical instability. Workspace analysis identifies reachable positions and poses, while manipulability analysis supports posture selection, trajectory planning, and robot design.

Effective professional practice combines mathematical modelling with calibration, unit control, joint-limit handling, collision checking, singularity avoidance, and validated safety systems.

Key Terms

Configuration: Complete set of robot joint variables.

Pose: Position and orientation of a rigid body.

Forward kinematics: Calculation of pose from joint variables.

Inverse kinematics: Calculation of joint variables from a desired pose.

Reference frame: Coordinate system attached to a robot component or environment.

Rotation matrix: Orthogonal 3 × 3 matrix representing orientation.

Homogeneous transformation: 4 × 4 matrix combining rotation and translation.

Rigid-body transformation: Motion preserving distances and angles.

Denavit–Hartenberg convention: Standard parameterization of serial-link geometry.

Link length: Distance between adjacent joint axes measured along a DH x-axis.

Link twist: Angle between adjacent joint axes measured about a DH x-axis.

Joint offset: Translation along a DH joint axis.

Jacobian: Matrix mapping joint rates to Cartesian velocity.

Twist: Combined linear and angular velocity vector.

Singularity: Configuration in which the Jacobian loses rank.

Workspace: Set of positions or poses achievable by a robot.

Manipulability: Measure of the robot’s ability to generate Cartesian motion.

Condition number: Ratio of largest to smallest singular value.

Calibration: Estimation or correction of model parameters using measurements.

Review Questions

Short Answer Questions

1. What is the difference between robot configuration and end-effector pose?
2. Why must the order of homogeneous transformation multiplication be preserved?
3. State the four standard Denavit–Hartenberg parameters.
4. What physical information is represented by the columns of a Jacobian?
5. Why can an inverse-kinematic problem have more than one solution?

Long Answer Questions

1. Explain the construction and use of homogeneous transformations for a serial robot.
2. Describe the standard Denavit–Hartenberg frame-assignment procedure and discuss the consequences of mixing standard and modified conventions.
3. Derive the forward kinematics and planar Jacobian of a two-revolute-joint arm.
4. Explain the causes and practical effects of robot singularities.
5. Compare rotation matrices, roll–pitch–yaw angles, axis–angle coordinates, and quaternions for representing orientation.

Analytical / Application Questions

1. A two-link planar robot has l₁ = 0.5 m and l₂ = 0.3 m. Determine whether the target point (0.9 m, 0 m) is reachable, and justify the result using workspace limits.
2. For a two-link planar arm, analyse the configuration θ₂ = 0. Explain why the Jacobian becomes singular and propose one practical trajectory-planning response.
3. Design a kinematic validation procedure for a pick-and-place robot that includes frame verification, forward-kinematic testing, inverse-kinematic branch selection, Jacobian monitoring, and safety checks.

Multiple Choice Questions with Answers

Question 1

Which matrix property is required for a valid 3 × 3 rotation matrix?

A. RᵀR = 0

B. RᵀR = I and det(R) = +1

C. det(R) = 0

D. R must be symmetric

Correct Answer: B. RᵀR = I and det(R) = +1

Question 2

In the standard DH convention, which parameter is normally variable for a revolute joint?

A. Link length aᵢ

B. Link twist αᵢ

C. Joint angle θᵢ

D. Link offset dᵢ

Correct Answer: C. Joint angle θᵢ

Question 3

What does the upper-right 3 × 1 block of a homogeneous transformation represent?

A. The rotation axis

B. The translation position vector

C. The determinant

D. The joint velocity vector

Correct Answer: B. The translation position vector

Question 4

For a prismatic joint, which Jacobian column components are correct?

A. Jᵥᵢ = zᵢ₋₁ and Jωᵢ = 0

B. Jᵥᵢ = 0 and Jωᵢ = zᵢ₋₁

C. Both components are zero

D. Jᵥᵢ = zᵢ₋₁ × (oₙ − oᵢ₋₁) and Jωᵢ = zᵢ₋₁

Correct Answer: A. Jᵥᵢ = zᵢ₋₁ and Jωᵢ = 0

Question 5

For a two-link planar arm, when does det(J) = l₁l₂ sinθ₂ become zero?

A. θ₂ = 30°

B. θ₂ = 45°

C. θ₂ = 90°

D. θ₂ = 0° or 180°

Correct Answer: D. θ₂ = 0° or 180°

Question 6

A target point for a two-link arm is unreachable in the basic inverse-kinematic formula when the computed value of cosθ₂ is:

A. Between −1 and 1

B. Exactly 0

C. Greater than 1 or less than −1

D. Equal to 0.5

Correct Answer: C. Greater than 1 or less than −1

Question 7

What does a robot singularity indicate?

A. The robot has maximum payload capacity

B. The Jacobian has lost rank

C. All joints are stationary permanently

D. The transformation matrix has no translation

Correct Answer: B. The Jacobian has lost rank

Question 8

Which measure is associated with Yoshikawa velocity manipulability?

A. √det(JJᵀ)

B. det(J) + 1

C. trace(J) only

D. 1 / det(R)

Correct Answer: A. √det(JJᵀ)

Question 9

For an unrestricted two-link planar arm, the maximum radial reach is:

A. |l₁ − l₂|

B. l₁l₂

C. l₁ + l₂

D. l₁² + l₂²

Correct Answer: C. l₁ + l₂

Question 10

What is the principal advantage of damped least-squares inverse kinematics near a singularity?

A. It eliminates all position error

B. It reduces velocity amplification and numerical instability

C. It guarantees every target is reachable

D. It removes the need for calibration

Correct Answer: B. It reduces velocity amplification and numerical instability