Page 1 – Robotics Fundamentals & Control Architectures

1.1 Engineering Domains Using Robotics

• Manufacturing robots (arms, manipulators, AGVs)
• Civil inspection robots (bridge scanning, tunnel mapping)
• Mechanical systems automation (assembly, welding, machining)
• Electrical inspection drones (powerlines, substations)
• Autonomous vehicles and UAVs

1.2 Kinematic Modelling

Robotics relies heavily on kinematics:

• Forward kinematics: joint → end-effector pose
• Inverse kinematics: pose → joint configuration
• Jacobian matrices: velocity mapping & singularities
• Denavit-Hartenberg parameters: systematic link representation

1.3 Dynamic Modelling

Robot dynamics follow nonlinear, coupled equations:

• Lagrangian formulation
• Newton–Euler mechanics
• Inertial matrices, Coriolis forces, centrifugal terms
• Friction, backlash, actuator saturation

1.4 Control Architecture Layers

• Low-level control – motor torque, velocity loops
• Mid-level control – trajectory tracking, servo control
• High-level planning – task sequencing, navigation, perception

1.5 Challenges in Robotics Control

• Nonlinear dynamics and unmodelled disturbances
• Time delays and communication latency
• Complex environments and dynamic obstacles
• Stability constraints under learning-based control

Page 2 – Classical Control: PID, MPC, State-Space Models

2.1 PID Control

A foundational controller in engineering systems:

• Proportional – error magnitude
• Integral – accumulated error
• Derivative – rate of error change

2.2 Limitations of PID

• Not adaptable to nonlinearity
• Sensitive to disturbances and friction
• Cannot easily handle constraints

2.3 State-Space Modelling

Defines system behaviour through equations of state:

• x'(t) = Ax + Bu (linear)
• y(t) = Cx + Du
• Useful for MIMO systems (multi-input, multi-output)
• Allows controllability & observability analysis

2.4 Observers and Estimators

• Kalman Filter & Extended Kalman Filter (EKF)
• Unscented Kalman Filter (UKF)
• Luenberger observer
• State estimation for robots under sensor noise

2.5 Model Predictive Control (MPC)

• Optimises control actions over a prediction horizon
• Handles constraints naturally
• Suitable for robotic arms, UAVs, HVAC systems, motors
• AI surrogates accelerate MPC computations

Page 3 – Reinforcement Learning (RL) for Engineering Systems

3.1 Overview of RL for Control

Reinforcement Learning (RL) learns control policies through interaction, useful for complex, nonlinear engineering systems where analytical control laws are insufficient.

3.2 RL Components

• State (s): system condition (joint angles, velocities, position)
• Action (a): torque, velocity, steering command
• Reward (r): performance objective
• Policy (π): mapping from states to actions

3.3 Common RL Algorithms for Robotics

• DQN – discrete action control, simple tasks
• DDPG – continuous control for robotic manipulators
• PPO – stable and widely used in industrial applications
• SAC – entropy-regularised for exploration robustness

3.4 RL for Robotic Manipulation

• Grasping objects with uncertain shapes
• Learning optimal torque profiles
• Trajectory optimisation under joint limits
• Cooperative multi-arm manipulation

3.5 RL for Mobile Robotics & UAVs

• Navigation through cluttered environments
• Obstacle avoidance under dynamic conditions
• Path optimisation under wind or turbulence
• Energy-minimised trajectory generation

3.6 RL Training Challenges

• Sample inefficiency
• Instability & divergence
• Reality gap (simulation → real-world)
• Safety-critical constraints

Page 4 – Hybrid AI–Control Systems & Autonomous Engineering

4.1 Combining Classical Control with AI

AI augments classical control, rather than replacing it. Hybrid frameworks merge stability guarantees with learning-based adaptability:

• PID loop with AI gain tuning
• MPC using AI surrogates for fast predictions
• RL policies bounded by control-theoretic constraints

4.2 Perception–Planning–Control Pipeline

• Perception – vision, LiDAR, thermal, ultrasonic
• Planning – trajectory generation, obstacle avoidance
• Control – low-level execution of planned actions

4.3 Vision-Based Robotic Control

• CNN-based object detection for grasping
• Pose estimation via deep learning
• Optical flow for UAV attitude corrections
• Semantic mapping for navigation

4.4 Digital Twins for Robotics

• Sim-to-real transfer
• Motion planning under structural constraints
• Rapid policy training in virtual environments
• Fault-tolerant control validation

4.5 Autonomous Systems

• Self-driving vehicles
• Warehouse automation
• Construction robotics (3D printing, inspection)
• Distributed robotic swarms

Page 5 – Safety, Certification, Ethics & Deployment in Engineering Robotics

5.1 Safety in Robotic Systems

• ISO 10218 – industrial robot safety
• Sensor redundancy for collision avoidance
• Fail-safe behaviours
• Geofencing and virtual boundaries
• Emergency stop thresholds

5.2 Certification of AI-Enabled Control Systems

• Performance validation under disturbances
• Verification of RL policy stability
• Testing for off-nominal cases
• Robustness across hardware variations

5.3 Ethical Engineering Considerations

• Transparency in autonomous decision-making
• Bias reduction in perception systems
• Human–robot collaboration ethics
• Responsibility for failures and accidents

5.4 Deployment Considerations

• Real-time compute budget and latency constraints
• Industrial network integration
• Hardware-in-the-loop testing (HIL)
• Model and controller version control

5.5 Summary

Module 4 explored robotics, classical control, reinforcement learning, hybrid AI–control systems, and safety frameworks essential for deploying autonomous engineering systems. These methods underpin modern manufacturing, inspection, mobility, and industrial automation.

Module 5 will examine AI Governance, Engineering Ethics & Industry 5.0.