4. Robot Dynamics and Mechanical Design

Introduction and Context

Robot motion is governed not only by geometry and kinematics but also by force, torque, mass distribution, stiffness, friction, and energy. Kinematic analysis answers where a robot is and how it moves; dynamic analysis explains why it moves in a particular way and what actuator effort is required. Mechanical design determines whether the resulting motion can be achieved safely, accurately, repeatedly, and economically.

For a postgraduate engineer, robot dynamics is therefore more than the derivation of equations of motion. It is a design discipline linking mechanism architecture, actuator sizing, transmission selection, structural design, control performance, payload capability, vibration, and energy consumption. A robot may possess adequate reach and nominal positioning accuracy yet fail in operation because its joints saturate, its links deflect, its natural frequencies lie within the operating range, or its thermal capacity is insufficient.

This chapter develops the principal methods used to model robot dynamics, with emphasis on Newton–Euler and Lagrangian formulations. It then connects these models with mechanical design decisions involving links, joints, payload, stiffness, vibration, accuracy, and dynamic performance.

Learning Objectives

After studying this chapter, the reader should be able to:

1. Distinguish kinematic, dynamic, static, and quasi-static robot models.

2. Formulate robot equations of motion using Newton–Euler and Lagrangian methods.

3. Interpret inertia, Coriolis, centrifugal, gravity, friction, and external-load terms.

4. Estimate actuator torque, power, payload capability, and energy demand.

5. Evaluate link and joint designs with respect to stiffness, strength, mass, wear, and backlash.

6. Explain how structural flexibility and vibration affect accuracy and control.

7. Select suitable design and validation methods for industrial robotic mechanisms.

8. Apply dynamic reasoning to an integrated robot-design problem.

From Motion Description to Force Requirement

For an n-degree-of-freedom robot, the joint-coordinate vector is written as

q = [q₁, q₂, …, qₙ]ᵀ.

The velocity and acceleration vectors are q̇ and q̈. A kinematic model maps q, q̇, and q̈ into the position, velocity, and acceleration of links and end effectors. A dynamic model maps these quantities, together with loads and physical properties, into the actuator forces or torques required to produce motion.

For a rigid-link robot with concentrated joint actuation, the standard equation is

M(q)q̈ + C(q,q̇)q̇ + g(q) + f(q̇) + τ_ext = τ,

where M(q) is the positive-definite joint-space inertia matrix, C(q,q̇)q̇ represents Coriolis and centrifugal effects, g(q) is the gravity vector, f(q̇) represents friction and other dissipative effects, τ_ext is the generalized load caused by external forces, and τ is the actuator torque vector.

The equation is often written more compactly as

τ = M(q)q̈ + h(q,q̇) + g(q) + f(q̇) + τ_ext.

The distinction between model terms is important. Inertia effects depend on acceleration; gravity depends primarily on configuration; Coriolis and centrifugal effects depend on joint motion; friction depends on relative motion, preload, temperature, and transmission condition. A motor selected only from the maximum static torque may be unable to deliver the required acceleration or withstand transient loads.

Newton–Euler Formulation: Force and Moment Balance

The Newton–Euler method applies Newton’s second law to the translation of each link and Euler’s rotational equation to its rotation. For link i, expressed in a suitable reference frame,

Fᵢ = mᵢaᵢ,

Nᵢ = Iᵢαᵢ + ωᵢ × (Iᵢωᵢ),

where Fᵢ is the resultant force, mᵢ is link mass, aᵢ is the acceleration of its centre of mass, Nᵢ is the resultant moment about the centre of mass, Iᵢ is the inertia tensor, αᵢ is angular acceleration, and ωᵢ is angular velocity.

In a serial manipulator, the method is usually implemented through a forward recursion followed by a backward recursion. The forward pass computes angular velocity, angular acceleration, and centre-of-mass acceleration from the base toward the end effector. The backward pass transfers forces and moments from the end effector toward the base, adding the effects of link inertia, gravity, and external loads. The resulting joint torque is obtained by projecting the net joint moment onto the joint axis for a revolute joint, or the net force onto the joint direction for a prismatic joint.

The Newton–Euler method is physically transparent and computationally efficient. Its recursive form is especially valuable for real-time control, where calculating inverse dynamics repeatedly is necessary. It also provides direct insight into bearing reactions, joint loads, and structural forces. Its main difficulty is bookkeeping: coordinate frames, signs, cross products, and the location of centres of mass must be handled consistently.

Lagrangian Formulation: Energy-Based Modelling

The Lagrangian approach begins with the kinetic and potential energies of the robot. The Lagrangian is

L(q,q̇) = T(q,q̇) − V(q),

where T is total kinetic energy and V is potential energy. For each generalized coordinate qᵢ, the equation is

d/dt(∂L/∂q̇ᵢ) − ∂L/∂qᵢ = τᵢ − τᵢ,loss + Qᵢ,ext.

Here τᵢ,loss represents dissipative effects and Qᵢ,ext is a generalized external force. For a rigid robot, kinetic energy includes translational and rotational contributions:

T = Σᵢ [½mᵢvᵢᵀvᵢ + ½ωᵢᵀIᵢωᵢ],

while gravitational potential energy is commonly expressed as

V = Σᵢ mᵢgᵀrᵢ,

with sign depending on the chosen gravity vector and coordinate convention.

The Lagrangian method produces the same physical equations as Newton–Euler analysis when the assumptions and parameters are identical. It is often more convenient for symbolic derivation, theoretical analysis, energy-based control, and systems containing coupled coordinates. It can become algebraically cumbersome for manipulators with many links, particularly when friction, gear elasticity, cable routing, and contact constraints must be included explicitly.

A useful structural property is that the inertia matrix is symmetric and positive definite. Furthermore, with an appropriate definition of the Coriolis matrix, the matrix combination Ṁ − 2C is skew-symmetric. This property is important in stability proofs for robot control and reflects conservation relationships in the mechanical model.

Static, Quasi-Static, and Dynamic Loads

Static analysis assumes q̇ = 0 and q̈ = 0. The required actuator effort then primarily balances gravity and external loads. Static torque is useful for holding a payload, but it is not a complete sizing criterion.

Quasi-static analysis neglects inertial effects when motion is slow relative to the mechanical time scale. It is appropriate for slow assembly, inspection, or positioning operations, provided that acceleration and deceleration loads are genuinely small.

Dynamic analysis includes acceleration, velocity-dependent coupling, friction, impacts, and changing external loads. The peak torque may occur during a rapid reversal rather than at the configuration with the largest gravitational moment. RMS torque is also important because it determines heating over a duty cycle. A motor may tolerate a short peak torque but fail thermally if its continuous or RMS demand is excessive.

Mechanical power at a revolute joint is

P = τq̇,

and for a prismatic joint it is

P = Fv.

Regenerative braking can return some kinetic energy to the electrical system, but transmission losses, drive limitations, and energy-storage architecture determine how much energy is actually recovered.

Mechanical Design of Links and Joints

Link architecture

Robot links should provide the required stiffness and strength with minimum moving mass. Reducing distal mass is particularly valuable because it lowers the inertia reflected to upstream joints. Hollow beams, ribbed castings, welded box sections, carbon-fibre structures, and topology-optimized components are common approaches. The selected design must also support cable routing, maintainability, sealing, manufacturing tolerances, and attachment of sensors and tooling.

Strength design checks yield stress, fatigue, buckling, bearing stress, and local stress concentration. Stiffness design checks deflection and twist under payload, gravity, acceleration, and process forces. In precision robots, stiffness often governs before material strength does. A link that does not yield may nevertheless deflect enough to violate the position tolerance.

For a simple cantilever carrying a transverse force F at its end, the elastic tip deflection is approximately

δ = FL³/(3EI),

where L is length, E is Young’s modulus, and I is the second moment of area. The cubic dependence on length demonstrates why long reach is mechanically expensive. Increasing section depth can raise I substantially, whereas simply adding material near the neutral axis is less effective.

Joint and transmission design

Joints must transmit torque while controlling friction, backlash, compliance, heat, and wear. Common transmission elements include harmonic drives, planetary gearboxes, cycloidal reducers, belt drives, ball screws, rack-and-pinion systems, and direct-drive motors.

Gear reduction increases output torque and can reduce motor size, but it introduces efficiency losses, reflected inertia, backlash or torsional compliance, and possibly limited backdrivability. Direct drive eliminates reduction error and can provide excellent force transparency, but the motor generally becomes larger and heavier. Harmonic drives offer high reduction in a compact package, but their flexsplines and bearings require careful evaluation for fatigue and torsional stiffness.

Bearing selection considers radial load, axial load, overturning moment, speed, life, lubrication, contamination, and preload. Joint seals protect the mechanism but increase friction. Cable harnesses can generate configuration-dependent torques and should not be ignored in high-accuracy systems.

Payload and workspace

Payload is not a single universal number. It depends on posture, acceleration, tool centre-point location, allowable deflection, joint torque, thermal duty, and required accuracy. A payload mounted far from the wrist produces a larger moment than the same mass close to the joint. The effective load includes the tool, workpiece, gripper, cables, and dynamic amplification.

A useful preliminary check for a horizontal revolute joint is

τ ≈ (m_p g l) + τ_acc,

where m_p is payload mass, l is the perpendicular distance from the joint axis, and τ_acc approximates the inertial contribution. This estimate must later be replaced by a full multibody dynamic model.

Stiffness, Compliance, Accuracy, and Repeatability

Robot accuracy is influenced by geometric calibration, encoder resolution, transmission error, thermal expansion, structural deflection, joint compliance, friction, and control error. Repeatability describes the ability to return to the same location under similar conditions; accuracy describes closeness to the commanded or calibrated location. A robot can be highly repeatable but systematically inaccurate.

For small deflections, Cartesian stiffness may be approximated by

K_x = (J K_q⁻¹ Jᵀ)⁻¹,

where J is the Jacobian and K_q is the joint stiffness matrix. The relation between an external Cartesian wrench F_x and displacement Δx is approximately

F_x = K_xΔx.

This expression shows that stiffness varies with posture. Near singular configurations, some Cartesian directions may become weak even when individual joints are stiff. Compliance can be beneficial in assembly and force control, but uncontrolled compliance causes contour error, chatter, and settling delay.

Error budgeting should separate systematic errors, such as link-length deviations, from random errors, such as encoder noise and friction variation. Calibration can reduce systematic geometric error but cannot fully remove load-dependent deflection or temperature-dependent changes.

Vibration and Dynamic Performance

A robot is a distributed mechanical system with multiple natural frequencies and mode shapes. A simplified single-degree-of-freedom model is

mẍ + cẋ + kx = F(t),

with natural frequency

ω_n = √(k/m)

and damping ratio

ζ = c/(2√(km)).

If excitation frequency approaches a natural frequency, resonance may produce large displacement and acceleration. Excitation can arise from periodic joint motion, gear-mesh frequency, servo sampling, payload changes, impacts, or process forces. The first flexible mode is often the limiting factor in fast pick-and-place operation.

Increasing stiffness generally raises natural frequency, while reducing moving mass can do the same. Damping reduces resonant amplification but may increase friction and heat. Structural redesign, trajectory shaping, input filtering, notch filters, and active vibration control can be combined. A rigid-body model is inadequate when flexible modes lie within the control bandwidth or when settling time is dominated by structural oscillation.

Dynamic performance is evaluated through peak and RMS torque, maximum speed and acceleration, settling time, contour error, vibration amplitude, thermal load, energy consumption, and cycle time. These measures must be assessed over the complete duty cycle rather than at one nominal posture.

A Practical Design and Validation Workflow

A disciplined design process begins with requirements: payload, reach, workspace, cycle time, accuracy, repeatability, process forces, environmental conditions, safety category, and expected service life. The designer then selects a mechanism architecture and establishes mass, inertia, centre-of-mass, and transmission assumptions.

The next stage is inverse dynamic simulation over representative trajectories. The simulation should include gravity, payload variation, acceleration limits, friction estimates, transmission efficiency, and external forces. Worst-case torque and speed are compared with motor, gearbox, bearing, and brake ratings. RMS torque and thermal models are used for continuous-duty assessment.

Finite-element analysis evaluates stress, deflection, buckling, and modal frequencies. Multibody simulation helps identify joint reactions, flexible-body interaction, and transient loads. Prototype testing should measure torque, temperature, vibration, backlash, compliance, and end-effector error. Correlation between model and test is essential; unexplained discrepancies often arise from omitted cable forces, bearing preload, friction, or joint flexibility.

Design margins should reflect uncertainty rather than simply multiplying every quantity by an arbitrary factor. Excessive margin increases mass and inertia, which can reduce dynamic performance and create a self-reinforcing design penalty.

Illustrative Case Study: High-Speed Palletizing Robot

Background

Consider an illustrative four-axis palletizing robot required to move a 20 kg product using a 12 kg gripper. The product must be placed at several pallet locations within a short cycle time. The robot has a long second link and a wrist carrying the gripper and product.

Situation and problem

A preliminary design meets the reach requirement and holds the payload under static conditions. During trial operation, the shoulder motor reaches its current limit during acceleration, and the end effector oscillates after sudden stops. Placement repeatability is acceptable at low speed but deteriorates at the required cycle time.

Analysis

A static calculation underestimated the shoulder torque because it neglected angular acceleration of the long link and the reflected inertia of the distal payload. In addition, the long link has low bending stiffness, and its first bending mode lies close to the dominant excitation produced by the trapezoidal velocity profile. The gearbox has adequate nominal torque but insufficient torsional stiffness for the desired settling time.

An inverse-dynamics model is created using measured masses and centres of mass. Newton–Euler recursion identifies the shoulder and elbow as the dominant torque contributors. An energy-based model confirms the gravity and inertial terms. Finite-element analysis predicts excessive tip deflection and a flexible mode near the operating excitation band. A duty-cycle calculation shows that RMS torque, not only peak torque, is thermally critical.

Approach and solution

The design team reduces distal mass by replacing a solid wrist housing with a ribbed hollow structure and relocates a drive component closer to the base. The second link is redesigned with a deeper closed section, increasing its bending stiffness without a proportional mass increase. A gearbox with higher torsional stiffness is selected, and the trajectory is modified to use jerk-limited acceleration rather than abrupt changes.

The servo controller incorporates feedforward torque based on the inverse dynamic model. A notch filter is introduced only after modal testing confirms the relevant resonance; excessive filtering is avoided because it would reduce control bandwidth. The payload range is divided into validated operating envelopes rather than represented by one nominal value.

Outcome

The redesigned mechanism reduces peak shoulder torque, raises the first flexible-mode frequency, and shortens settling time. The important result is not merely a larger motor. The improvement comes from coordinated changes to mass distribution, structural stiffness, transmission compliance, trajectory design, and model-based control.

Learning points

The case demonstrates four general principles. First, payload capacity is posture- and trajectory-dependent. Second, stiffness and mass distribution influence both accuracy and actuator sizing. Third, vibration is a coupled mechanical-control problem. Finally, dynamic validation must use the actual duty cycle and include transient behaviour.

Challenges, Risks, and Common Design Errors

A frequent error is to size actuators from static gravity torque alone. Other errors include treating payload as a point mass at the wrong location, ignoring tool and cable mass, using catalogue torque without checking speed and duty cycle, neglecting gearbox efficiency, and assuming that encoder resolution guarantees Cartesian accuracy.

Model uncertainty is unavoidable. Friction changes with temperature and speed; joint stiffness may be nonlinear; bearings and gear teeth introduce manufacturing variation; and contact events can generate loads that are difficult to predict. Conservative assumptions should be applied to safety-critical conditions, but validation testing is preferable to arbitrary overdesign.

Backlash and compliance are sometimes confused. Backlash is lost motion or dead zone, while compliance is elastic deformation under load. Both affect accuracy, but they require different remedies. Preload can reduce backlash but increase friction and bearing load. Increasing controller gain may reduce tracking error in a rigid system but can excite flexible modes in a compliant mechanism.

Safety, Ethics, and Professional Responsibility

Dynamic design is inseparable from safety. Stored energy in elevated links, brakes, gear trains, and payloads must be considered during power loss and emergency stopping. Risk assessment should address overspeed, uncontrolled release, collision, pinch points, unexpected restart, and failure of sensing or braking systems. Safety factors and protective devices should be justified by applicable industrial standards and the intended operating environment.

Engineers must not report payload or accuracy values without stating the conditions under which they were measured. Misleading performance claims can produce unsafe integration decisions. Traceable calculations, configuration control, validation records, and clear operating envelopes are part of professional engineering practice.

Digital Tools and Emerging Directions

Modern robot development combines symbolic dynamics, numerical multibody simulation, finite-element analysis, digital twins, and experimental system identification. Automatic differentiation and recursive algorithms can generate dynamic models suitable for optimization and real-time control. Data from current sensors, encoders, accelerometers, and thermal sensors can support condition monitoring and predictive maintenance.

Machine-learning methods are increasingly used to estimate friction, payload, and unmodeled dynamics. Their use should remain bounded by physically meaningful models and validated operating conditions. A learned model that performs well in one payload or temperature range may fail under extrapolation. Digital twins are most valuable when continuously correlated with measured data rather than treated as static visualizations.

Emerging mechanical designs include lightweight composite links, variable-stiffness actuators, series elastic actuators, torque-controlled transmissions, and integrated motor-drive-sensor modules. These technologies can improve force control and safety, but they introduce bandwidth, calibration, thermal, and reliability challenges.

Best Practices for Engineering Practice

A robust robot-dynamics workflow should:

• define payload, posture, trajectory, and duty-cycle assumptions explicitly;
• use consistent frames, units, and sign conventions;
• retain mass and inertia properties from CAD but verify them experimentally when critical;
• evaluate peak, continuous, and RMS actuator demand;
• include transmission efficiency, friction, cable loads, and external process forces;
• assess stiffness and modal behaviour together with strength;
• validate rigid-body models against measurements before relying on them for control;
• separate accuracy, repeatability, resolution, and compliance in specifications;
• design for safe stopping and foreseeable single faults; and
• document the operating envelope and uncertainty of every key performance claim.

Chapter Summary

Robot dynamics connects motion requirements with the forces, torques, power, and energy demanded from a mechanical system. The Newton–Euler method derives these requirements through recursive force and moment balances, whereas the Lagrangian method derives them from kinetic and potential energy. Both lead to the standard robot equation involving inertia, velocity coupling, gravity, friction, and external loads.

Mechanical design must consider more than static strength. Link mass and inertia influence actuator sizing; joint transmissions affect efficiency, backlash, compliance, and backdrivability; payload capability varies with posture and acceleration; and stiffness determines load-dependent accuracy. Flexible modes and vibration can limit cycle time even when motors have adequate torque. A successful design therefore integrates dynamic simulation, structural and modal analysis, thermal evaluation, control design, prototype testing, and safety assessment.

Key Terms

Inverse dynamics: Calculation of actuator forces or torques required for a specified motion.

Forward dynamics: Calculation of motion resulting from specified actuator forces or torques.

Inertia matrix: Configuration-dependent matrix relating joint accelerations to inertial torques.

Coriolis effect: Velocity-dependent coupling associated with motion in multiple coordinates.

Centrifugal effect: Velocity-dependent inertial effect directed relative to rotating motion.

Generalized force: Force or torque associated with a generalized coordinate.

Lagrangian: Difference between a system’s kinetic and potential energies.

Newton–Euler method: Dynamic formulation based on translational and rotational equilibrium.

Payload: Mass carried by the robot, including tooling and workpiece where specified.

Reflected inertia: Inertia viewed at a shaft or joint after transmission transformation.

Backlash: Lost motion caused by clearance in a transmission or joint.

Compliance: Elastic displacement produced by an applied load.

Cartesian stiffness: Resistance of the robot to end-effector displacement under a Cartesian load.

Natural frequency: Frequency at which a system tends to oscillate freely.

Damping ratio: Dimensionless measure of damping relative to critical damping.

RMS torque: Root-mean-square torque over a specified duty cycle.

Repeatability: Ability to return to the same location under repeated conditions.

Accuracy: Closeness of the achieved position to the desired or reference position.

Modal analysis: Determination of natural frequencies and mode shapes.

Duty cycle: Time sequence of operating loads, speeds, and rest periods.

Review Questions

Short Answer Questions

1. What physical effects are represented by the terms M(q)q̈ and g(q) in a robot dynamic equation?

2. State one important advantage and one limitation of the Newton–Euler method.

3. Distinguish robot accuracy from repeatability.

4. Why is RMS torque important in actuator selection?

5. What is the relationship between structural stiffness and load-dependent positioning error?

Long Answer Questions

1. Derive the Lagrange equation for a multi-degree-of-freedom robot and explain the physical meaning of each term.

2. Compare Newton–Euler and Lagrangian formulations in terms of derivation, computation, interpretation, and control applications.

3. Explain how link mass distribution, transmission selection, and payload location affect dynamic performance.

4. Discuss the causes of vibration in industrial robots and describe mechanical and control-based methods for reducing it.

5. Develop a complete workflow for validating the mechanical and dynamic design of a new robot manipulator.

Analytical / Application Questions

1. A payload is moved twice as far from a revolute joint while its mass remains constant. Explain the effects on gravity torque, structural deflection, and likely actuator requirements.

2. A robot has sufficient peak torque but overheats during repeated operation. Identify the likely causes and propose a calculation and testing strategy.

3. A manipulator is highly repeatable at low speed but inaccurate under changing payloads. Diagnose the dominant error sources and recommend mechanical, calibration, and control remedies.

22. Multiple Choice Questions with Answers

Question 1

Which term in the standard rigid-robot dynamic equation primarily represents configuration-dependent inertial effects?

A. g(q)

B. M(q)q̈

C. f(q̇)

D. τext

Correct Answer: B. M(q)q̈

Question 2

In the Newton–Euler method, the backward recursion is mainly used to compute:

A. Joint positions from actuator speeds

B. Link masses from CAD geometry

C. Forces and moments transmitted toward the base

D. Natural frequencies from damping ratios

Correct Answer: C. Forces and moments transmitted toward the base

Question 3

Why can a robot with adequate static payload capacity still fail during rapid motion?

A. Static analysis always overestimates friction

B. Dynamic inertial and velocity-dependent torques may exceed actuator capacity

C. Gravity disappears during acceleration

D. Payload has no effect on joint torque during motion

Correct Answer: B. Dynamic inertial and velocity-dependent torques may exceed actuator capacity

Question 4

For a cantilever link carrying an end force, the approximate tip deflection varies with link length as:

A. L

B. L²

C. L³

D. 1/L

Correct Answer: C. L³

Question 5

Which statement correctly distinguishes accuracy from repeatability?

A. Accuracy concerns speed; repeatability concerns torque

B. Accuracy is closeness to the target; repeatability is consistency of repeated results

C. Accuracy applies only to rotational joints

D. Repeatability can be improved only by increasing payload

Correct Answer: B. Accuracy is closeness to the target; repeatability is consistency of repeated results

Question 6

If a robot’s excitation frequency approaches one of its natural frequencies, the most likely result is:

A. Reduced inertial torque to zero

B. Resonant vibration and amplified displacement

C. Elimination of structural compliance

D. A guaranteed increase in accuracy

Correct Answer: B. Resonant vibration and amplified displacement

Question 7

Which change generally reduces the inertia reflected to an upstream joint?

A. Moving mass farther toward the end effector

B. Increasing distal component mass

C. Reducing distal mass and relocating components toward the base

D. Increasing cable slack without changing mass

Correct Answer: C. Reducing distal mass and relocating components toward the base

Question 8

A robot gearbox has adequate nominal torque but excessive torsional compliance. Which performance problem is most directly expected?

A. Improved settling time

B. Load-dependent positioning error and oscillation

C. Elimination of backlash

D. Reduced gravitational force

Correct Answer: B. Load-dependent positioning error and oscillation

Question 9

Why is RMS torque evaluated in addition to peak torque?

A. It estimates thermal loading over the duty cycle

B. It determines only encoder resolution

C. It removes the need for modal analysis

D. It is independent of operating time

Correct Answer: A. It estimates thermal loading over the duty cycle

Question 10

In the illustrative palletizing case, which combined strategy addressed both torque demand and vibration?

A. Using only a larger motor

B. Reducing distal mass, increasing stiffness, improving transmission stiffness, and shaping the trajectory

C. Removing all feedback sensors

D. Increasing payload while reducing link stiffness

Correct Answer: B. Reducing distal mass, increasing stiffness, improving transmission stiffness, and shaping the trajectory