9. Robotics and Autonomous Mechatronic Systems

Introduction and Context

Robotics is the engineering discipline concerned with machines that perceive their environment, make decisions, and generate physical action. It therefore represents a particularly complete expression of mechatronics: mechanical structures provide motion, actuators generate forces, sensors measure state and surroundings, control systems regulate behaviour, and software coordinates the entire system. Industrial robots, collaborative robots, autonomous mobile robots, surgical platforms, warehouse vehicles, and unmanned inspection systems all arise from this integration.

A robot is not merely a programmable manipulator. Its usefulness depends on the relationship between geometry, dynamics, sensing, computation, control, safety, and the task environment. A robot may possess excellent repeatability but poor absolute accuracy; a mobile platform may localise well in a laboratory but fail in a reflective industrial hall; and a vision system may identify an object correctly while still placing it incorrectly because the camera-to-robot transformation is inaccurate. Engineering decisions must therefore be made at system level.

This chapter develops the principal methods used to design and operate robotic and autonomous mechatronic systems. It connects rigid-body kinematics and dynamics with trajectory generation, machine vision, localisation, mapping, force control, and human–robot interaction. It also considers industrial deployment, safety, uncertainty, artificial intelligence, and the managerial implications of selecting and integrating robotic technology.

Learning Objectives

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

1. Model the kinematics and dynamics of robotic manipulators and mobile robots.

2. Distinguish position accuracy, repeatability, resolution, workspace, and payload capability.

3. Plan smooth, feasible, and collision-free robot trajectories.

4. Explain the roles of machine vision, calibration, localisation, mapping, and sensor fusion.

5. Select appropriate position, velocity, impedance, and force-control strategies.

6. Compare industrial robots, cobots, autonomous mobile robots, and autonomous machines.

7. Analyse safety, ethical, organisational, and implementation issues in robotic deployment.

8. Evaluate an autonomous robotic system using technical and operational performance measures.

Robotics as an Integrated Mechatronic System

A robotic system can be represented as a closed loop. Sensors measure joint position, velocity, force, torque, temperature, or environmental features. A state-estimation layer interprets these measurements. Planning software determines a desired action, while a controller computes actuator commands. The mechanical plant responds, and new measurements close the loop.

For a manipulator, the configuration is commonly described by the vector

q = [q₁, q₂, …, qₙ]ᵀ,

where each qᵢ is a revolute angle or prismatic displacement. The end-effector pose consists of position and orientation. Position may be represented by a vector p = [x, y, z]ᵀ, while orientation may be represented by a rotation matrix, Euler angles, or a unit quaternion. A homogeneous transformation matrix combines both:

T = [ R p ; 0 1 ],

where R is a 3 × 3 rotation matrix. The transformation describes the pose of one coordinate frame relative to another and permits successive link transformations to be multiplied.

This frame-based description is central to robot programming. The base frame, joint frames, tool frame, workpiece frame, camera frame, and world frame must be defined consistently. Many practical failures attributed to control are actually caused by incorrect frame conventions or poor calibration.

Manipulator Kinematics: From Joint Motion to Tool Pose

Forward kinematics determines the end-effector pose from known joint variables:

x = f(q).

For serial manipulators, the function f is obtained by multiplying link transformations. The Denavit–Hartenberg convention provides a systematic parameterisation using link length, link twist, link offset, and joint angle. Although alternative formulations are often preferred in modern software, the underlying principle remains the same: robot geometry maps configuration space into task space.

Inverse kinematics solves the reverse problem: given a desired pose x, determine q such that f(q) = x. A robot may have multiple solutions, no solution, or infinitely many solutions. Multiple solutions arise from configurations such as elbow-up and elbow-down postures. A practical solver must consider joint limits, self-collision, obstacle clearance, singularities, and continuity with the current configuration.

The differential relationship between joint velocity and end-effector velocity is

ẋ = J(q)q̇,

where J(q) is the Jacobian matrix. The Jacobian is useful for velocity control, singularity analysis, force transformation, and sensitivity assessment. Near a singular configuration, one or more task-space directions become difficult or impossible to generate. Joint velocities may become excessively large when the controller attempts to produce a modest Cartesian velocity. Damped least-squares inverse kinematics is often used to reduce this problem by trading exact tracking for numerical and physical robustness.

Important performance terms must be distinguished. Accuracy is the closeness of a realised pose to its commanded or nominal value. Repeatability is the ability to return to the same pose under repeated conditions. Resolution is the smallest detectable or commandable increment. A robot can be highly repeatable but inaccurate if its geometric calibration is biased. This distinction is important in assembly, machining, dispensing, and metrology.

Dynamics and Control of Motion

Kinematics describes motion without considering force. Dynamics relates motion to torque, inertia, gravity, friction, and external loads. A widely used manipulator model is

M(q)q̈ + C(q,q̇)q̇ + g(q) + τ_f(q̇) = τ + Jᵀ(q)F_ext,

where M is the inertia matrix, C represents Coriolis and centrifugal effects, g is the gravity vector, τ_f represents friction, τ is actuator torque, and F_ext is an external wrench applied at the end effector.

A basic joint-space controller uses proportional–derivative feedback:

τ = K_p(q_d − q) + K_d(q̇_d − q̇) + τ_ff,

where τ_ff may include gravity or dynamic compensation. High gains can improve tracking but may amplify sensor noise, excite structural flexibility, and increase contact forces. Model-based control can improve performance, but model uncertainty means that feedback remains essential.

Task-space control regulates position and orientation directly. It is appropriate when a tool must follow a Cartesian path, maintain a welding orientation, or exert a specified contact force. Hybrid position–force control separates task directions: position is controlled in unconstrained directions and force in constrained directions. Impedance control instead specifies a desired dynamic relationship between position error and interaction force. It is especially useful when the environment is uncertain because the robot behaves as though connected to the desired trajectory through virtual stiffness and damping.

Trajectory Planning and Execution

A robot does not simply move from one point to another; it must follow a time-dependent path while respecting limits on position, velocity, acceleration, jerk, torque, and collision risk. A geometric path specifies the sequence of configurations or poses. A trajectory adds timing.

Joint-space interpolation is computationally simple and guarantees joint-limit awareness when the interpolated values remain within limits, but the end-effector path may be unintuitive. Cartesian interpolation gives a straight or otherwise specified tool path, although inverse kinematics must be solved continuously and singularities may be encountered. A common smooth profile is a polynomial trajectory. Cubic polynomials ensure specified initial and final position and velocity; quintic polynomials additionally permit specified zero or finite acceleration at endpoints. S-curves limit jerk and reduce vibration, making them valuable for high-speed handling and delicate payloads.

Planning generally has two levels. Global planning identifies a collision-free route through the known environment, using methods such as graph search or sampling-based planners. Local planning adapts the motion to updated sensor information and moving obstacles. A valid plan must be checked against joint limits, actuator capabilities, dynamic constraints, tool orientation, and safety zones. Time-optimal motion is not necessarily production-optimal: excessive acceleration may increase wear, vibration, energy use, and rejected products.

Perception, Calibration, and Machine Vision

Autonomous action requires a robot to obtain information about objects and surroundings. Sensors may include encoders, cameras, depth cameras, laser scanners, force–torque sensors, inertial measurement units, and proximity devices. No sensor is universally superior; selection depends on range, resolution, lighting, material reflectivity, latency, cost, and environmental robustness.

Machine vision commonly involves image acquisition, preprocessing, segmentation, feature extraction, object recognition, pose estimation, and decision-making. Classical methods may use edges, contours, templates, or geometric features. Learning-based methods can classify complex objects and estimate poses from large data sets, but their performance depends on representative training data and appropriate validation.

Calibration establishes geometric relationships. Intrinsic calibration estimates camera parameters such as focal length and lens distortion. Extrinsic calibration estimates the transformation between the camera and a robot, workpiece, or world frame. Hand–eye calibration is particularly important in vision-guided manipulation. A calibration error can create systematic placement errors even when perception and robot control are individually functioning correctly.

Sensor fusion combines complementary information. For example, wheel odometry provides short-term motion estimates but drifts over distance; an inertial sensor responds rapidly but accumulates bias; and laser or vision observations provide environmental references but may be intermittent. Probabilistic filters, including extended Kalman filters and particle filters, combine measurements according to uncertainty models.

Localisation and Mapping for Autonomous Machines

A mobile robot must estimate its pose, commonly represented as position and orientation in a map. Localisation may be absolute, relative, or map-based. Dead reckoning integrates wheel or joint motion and is simple but accumulates error. Landmark-based localisation corrects drift using known features. Simultaneous localisation and mapping, or SLAM, addresses the coupled problem in which the robot builds a map while estimating its own pose.

A useful autonomy architecture separates mapping, global planning, local obstacle avoidance, and low-level control. The map may be a metric occupancy grid, a topological graph, a point cloud, or a semantic representation containing object categories. Global planning seeks a route to a destination; local planning reacts to people, vehicles, and newly detected obstacles.

Autonomy is not equivalent to the absence of human supervision. A system should define operating boundaries, confidence thresholds, degraded modes, emergency stops, and recovery procedures. When localisation confidence falls below an acceptable level, the robot may slow, stop, request assistance, or return to a safe location rather than continue with an unverified estimate.

Force Control and Physical Interaction

Position control is appropriate when the robot operates freely in space and accurate geometry is available. Contact tasks require consideration of force. Examples include insertion, polishing, deburring, grinding, assembly, and interaction with a human operator.

Force may be measured at the wrist, inferred from motor current, or estimated through deflection models. Direct force control attempts to regulate measured force, whereas impedance control regulates the apparent mechanical behaviour. Admittance control converts measured force into a motion command and can be useful when the robot has high mechanical stiffness but a compliant task is desired.

Contact stability depends on sampling rate, sensor noise, structural compliance, friction, delays, and environmental stiffness. A controller tuned in free space may become unstable during contact. Contact transitions should therefore be designed explicitly, with limits on force, velocity, and energy. Insertion tasks often benefit from compliance, search motions, chamfered tooling, and controlled alignment rather than from extremely accurate position control alone.

Industrial Robots, Cobots, and Autonomous Mobile Robots

Traditional industrial robots are designed for high speed, repeatability, payload, and operation within safeguarded workspaces. They are effective in welding, painting, palletising, machine tending, and high-volume assembly. Their deployment usually includes fencing, interlocked gates, safety-rated monitoring, and carefully defined operating modes.

Collaborative robots, or cobots, are designed to share space with people under specified conditions. Their value lies not simply in low speed, but in the combination of force limitation, monitored stopping, safety-rated sensing, suitable tooling, risk assessment, and task design. A cobot is not automatically safe for every application. Sharp tools, heavy payloads, pinch points, unexpected motion, and high stored energy may still create unacceptable risk.

Autonomous mobile robots transport materials, inspect facilities, or coordinate warehouse operations. Their effectiveness depends on navigation reliability, fleet management, traffic rules, charging strategy, communication coverage, and integration with manufacturing execution or warehouse management systems. An autonomous machine may combine a mobile base with a manipulator, vision, and task-level planning, creating a more complex verification and safety problem than either subsystem alone.

Human–Robot Interaction and System Governance

Human–robot interaction includes physical collaboration, supervisory control, programming by demonstration, natural-language instruction, and shared decision-making. Effective interaction requires predictable robot behaviour, clear indication of operating state, understandable alarms, and interfaces that reveal uncertainty rather than conceal it.

Programming by demonstration can reduce deployment time by allowing an operator to guide a robot through a task. However, demonstrated behaviour must be validated for repeatability, collision risk, hidden assumptions, and changes in product or environment. Natural-language interfaces can improve accessibility but should not be permitted to bypass safety logic or authorise unverified actions.

A professional deployment assigns responsibility for risk assessment, software updates, access control, data management, maintenance, and incident reporting. Safety standards and regional regulations should be consulted for the specific robot, application, machinery category, and workplace. Cybersecurity is also a safety concern: unauthorised access, corrupted trajectories, manipulated sensor data, or unavailable network services can produce physical consequences.

An Illustrative Industrial Case: Vision-Guided Collaborative Assembly

Background

A medium-sized manufacturer assembles small electromechanical modules in batches of varying product types. Previously, operators manually located housings, inserted components, and performed a visual check. The process was ergonomically repetitive and suffered from variation in component orientation.

Situation and Problem

Management proposed a collaborative robot with a camera, wrist force sensor, and automatic product identification. The proposed system had to handle several housing variants, work beside an operator, and avoid damaging connectors during insertion. A simple replacement of the operator by a position-controlled robot was rejected because component tolerances and fixture variation made rigid insertion unreliable.

Analysis

The engineering team decomposed the task into identification, grasping, orientation correction, insertion, verification, and handover. Vision was suitable for locating parts, but the camera required calibrated lighting and a rigid camera mount. The insertion stage required force feedback because visual pose accuracy alone could not guarantee alignment. A risk assessment identified pinch points, unexpected restart, sharp edges, dropped components, and interaction with the operator.

Approach and Solution

The robot used vision to estimate component pose and selected among prevalidated grasp configurations. A calibrated transformation converted camera coordinates into the robot base frame. Motion was planned in two stages: a fast collision-free approach followed by a slower compliant insertion. Impedance control allowed small lateral corrections while limiting insertion force. The operator remained responsible for replenishing components and initiating selected batches, while the robot reported confidence and stopped when visual or force conditions were outside validated ranges.

Before production, the team tested normal, abnormal, and foreseeable misuse cases. The system was integrated with a production database so that the correct robot program and inspection criteria were associated with each product code. Performance was measured using cycle time, first-pass yield, false detection rate, unplanned stops, near misses, and operator acceptance rather than cycle time alone.

Outcome

The illustrative system improved process consistency and reduced repetitive handling, but it did not eliminate human involvement. Operators became responsible for supervision, replenishment, exception recovery, and quality confirmation. The main technical lesson was that vision, force control, and safety logic had to be designed as one system. The main managerial lesson was that work redesign, training, maintenance ownership, and data governance were as important as the robot purchase.

Learning Points

1. Calibration errors can dominate the performance of an otherwise accurate robot.

2. Contact tasks often require compliance rather than greater positional stiffness.

3. Collaborative operation must be justified by application-specific risk assessment.

4. Deployment metrics should include safety, quality, reliability, and human factors.

5. A robot project is an organisational change project as well as an equipment project.

Challenges, Limitations, and Common Engineering Errors

Robotic performance is limited by uncertainty in geometry, sensing, friction, payload, timing, and the environment. Common errors include selecting a robot from payload alone, ignoring wrist inertia and reach, treating repeatability as accuracy, using an uncalibrated camera, planning only nominal paths, and testing only normal operation.

Autonomous systems can fail through perception ambiguity, environmental change, localisation loss, communication delay, software defects, or unexpected human behaviour. Learning-based perception may perform poorly on rare objects or lighting conditions not represented in training data. Simulation can accelerate development, but differences between simulated and physical environments must be identified and tested.

Maintenance is another frequent weakness. Encoders, cameras, force sensors, grippers, batteries, networks, and safety devices require inspection and calibration. A robot that achieves impressive performance during commissioning may become unreliable if preventive maintenance, spare parts, version control, and fault diagnosis are neglected.

Emerging Directions and Best Practices

Current development is moving toward more capable autonomous manipulation, multimodal perception, digital twins, cloud-connected fleet management, reinforcement learning, foundation models for task interpretation, and energy-aware planning. These technologies can improve flexibility, but they also increase the importance of validation, explainability, cybersecurity, and controlled deployment.

Sound practice begins with a clearly defined task and measurable acceptance criteria. The engineering team should then:

• analyse hazards and operational constraints before selecting hardware;
• choose sensors according to uncertainty, not novelty;
• maintain explicit coordinate frames and calibration records;
• use layered control, planning, monitoring, and safety functions;
• test nominal, boundary, fault, recovery, and misuse conditions;
• provide a safe degraded mode and a clear human override;
• record software, parameter, and model versions;
• evaluate total lifecycle cost, including integration and maintenance.

The most successful systems are usually not those with the most sophisticated algorithm, but those whose mechanical design, sensing, control, software, people, and operating procedures are coherently matched to the task.

Chapter Summary

Robotics integrates mechanical motion, sensing, computation, control, and interaction into a physical decision-making system. Manipulator kinematics relates joint variables to tool pose, while the Jacobian supports velocity and force analysis. Dynamics accounts for inertia, gravity, friction, actuator torque, and external interaction. Trajectory planning must satisfy geometric, kinematic, dynamic, and safety constraints.

Autonomous systems depend on calibrated perception, sensor fusion, localisation, mapping, planning, and feedback control. Position control is suitable for unconstrained motion, whereas force, impedance, or admittance control is valuable in contact tasks. Industrial robots prioritise speed and repeatability; cobots emphasise controlled human collaboration; and mobile robots require dependable localisation, navigation, fleet coordination, and recovery behaviour.

Effective deployment requires more than technical capability. Risk assessment, cybersecurity, governance, maintenance, workforce design, training, and lifecycle metrics determine whether a robotic system produces safe and sustainable value. The central engineering principle is systems integration under uncertainty.

Key Terms

Kinematics: Study of motion and geometry without considering forces.

Dynamics: Study of the relationship between forces, torques, mass, and motion.

Configuration space: Set of all permissible joint configurations of a robot.

Forward kinematics: Computation of end-effector pose from joint variables.

Inverse kinematics: Computation of joint variables required to achieve a desired pose.

Jacobian: Matrix relating joint velocities to end-effector velocities.

Singularity: Configuration in which one or more task-space motions become unavailable or ill-conditioned.

Homogeneous transformation: Matrix representation combining rotation and translation between coordinate frames.

Accuracy: Closeness of an achieved value to its desired or reference value.

Repeatability: Ability to return consistently to the same pose or result.

Trajectory: Time-parameterised motion consisting of a path and timing law.

SLAM: Simultaneous localisation and mapping by which a robot estimates its pose while constructing a map.

Sensor fusion: Combination of measurements from multiple sensors using their complementary information and uncertainties.

Impedance control: Control of the desired dynamic relationship between motion error and interaction force.

Admittance control: Conversion of measured force into a commanded motion response.

Hand–eye calibration: Estimation of the geometric transformation between a camera and a robot or tool.

Collaborative robot: Robot designed for specified forms of controlled operation in proximity to people.

Occupancy grid: Map representing the probability that spatial cells are occupied.

Digital twin: Computational representation linked to a physical system for monitoring, analysis, or control.

Review Questions

Short Answer Questions

1. Distinguish robot accuracy from repeatability.

2. What information is represented by a homogeneous transformation matrix?

3. Why can a robot encounter difficulty near a kinematic singularity?

4. State one reason force control is useful in insertion tasks.

5. What is the purpose of hand–eye calibration?

Long Answer Questions

1. Explain forward and inverse kinematics and discuss the practical constraints that affect inverse-kinematic solutions.

2. Describe the principal stages of trajectory planning for an industrial manipulator.

3. Explain how localisation, mapping, global planning, and local obstacle avoidance cooperate in an autonomous mobile robot.

4. Compare position control, hybrid position–force control, impedance control, and admittance control.

5. Discuss the technical, organisational, safety, and cybersecurity requirements for deploying a collaborative robot.

Analytical / Application Questions

1. A vision-guided robot repeatedly places parts in the wrong location by a nearly constant offset, although its repeatability is excellent. Analyse the likely causes and propose a verification procedure.

2. An autonomous warehouse vehicle performs well in a static test area but loses localisation when aisles become crowded and reflective. Recommend a sensing, mapping, and recovery strategy.

3. A robot must insert a connector into a housing with moderate dimensional variation while operating beside a worker. Design a control and safety architecture, and justify the principal performance measures.

22. Multiple Choice Questions with Answers

Question 1

Which relationship directly maps joint velocity to end-effector velocity?

A. x = f(q)

B. ẋ = J(q)q̇

C. M(q)q̈ + g(q) = 0

D. T = [R p; 0 1]

Correct Answer: B. ẋ = J(q)q̇

Question 2

A robot returns repeatedly to the same location but that location is offset from the commanded position. Which property is high and which is poor?

A. High accuracy and poor repeatability

B. Poor accuracy and high repeatability

C. Poor resolution and high accuracy

D. High workspace and poor payload

Correct Answer: B. Poor accuracy and high repeatability

Question 3

Why is impedance control particularly useful during uncertain contact tasks?

A. It eliminates the need for sensors

B. It guarantees zero contact force

C. It specifies a compliant dynamic relationship between motion and force

D. It prevents all singularities

Correct Answer: C. It specifies a compliant dynamic relationship between motion and force

Question 4

What is the principal purpose of hand–eye calibration?

A. To increase motor torque

B. To estimate the transformation between a camera and a robot or tool

C. To determine battery capacity

D. To generate a warehouse map without sensors

Correct Answer: B. To estimate the transformation between a camera and a robot or tool

Question 5

Which statement best describes SLAM?

A. It controls only the speed of a robot arm

B. It estimates robot pose while constructing a map

C. It converts force into motor current only

D. It calibrates lens distortion without motion

Correct Answer: B. It estimates robot pose while constructing a map

Question 6

A Cartesian straight-line tool path is required for welding. Which planning approach is most appropriate?

A. Unconstrained random joint motion

B. Cartesian path interpolation with continuous inverse-kinematic checking

C. Only endpoint positioning with no intermediate checks

D. Dead reckoning without a robot model

Correct Answer: B. Cartesian path interpolation with continuous inverse-kinematic checking

Question 7

Which situation most clearly requires force or compliance control rather than position control alone?

A. Moving an unloaded robot through free space

B. Returning a robot to a taught home position

C. Inserting a connector with dimensional variation

D. Reading a joint encoder

Correct Answer: C. Inserting a connector with dimensional variation

Question 8

Why might wheel odometry alone be inadequate for long-distance mobile-robot navigation?

A. It cannot measure joint angle

B. Its errors accumulate and cause drift

C. It always requires a camera calibration

D. It measures only external force

Correct Answer: B. Its errors accumulate and cause drift

Question 9

Which is the most appropriate interpretation of collaborative-robot safety?

A. A cobot is safe for every task because it operates slowly

B. Safety depends only on the robot brand

C. Safety must be established through application-specific risk assessment and protective measures

D. Safety is unnecessary when an operator is experienced

Correct Answer: C. Safety must be established through application-specific risk assessment and protective measures

Question 10

In the illustrative assembly case, why was a force sensor added to the vision-guided system?

A. To replace all calibration procedures

B. To permit compliant correction during connector insertion

C. To increase camera resolution

D. To create a global occupancy map

Correct Answer: B. To permit compliant correction during connector insertion