About this book:

This book provides a comprehensive engineering guide to building autonomous robotic agents by integrating high-level AI with resource-constrained embedded hardware. It frames the "perception-to-action stack" as a multi-layered architecture where raw sensor data is transformed into physical motion through a sequence of sensing, perception, cognition, and control. Central to the text is the requirement for strict real-time determinism; the authors argue that in robotics, an algorithm is only as good as its ability to meet hardware deadlines, minimize jitter, and maintain stability within millisecond-scale windows.

The technical core of the book details the transition from classical robotics to AI-driven autonomy. It covers foundational methods like Kalman and Particle filters for state estimation alongside modern deep learning approaches for visual SLAM and object detection. A significant portion of the text is dedicated to the optimization of these models for the "edge," exploring techniques such as quantization, pruning, and the use of specialized hardware accelerators like NPUs and FPGAs. By balancing traditional control theory (PID and Model Predictive Control) with Reinforcement Learning, the book provides a roadmap for creating agents that are both physically precise and behaviorally adaptive.

Beyond algorithmic design, the book addresses the operational realities of deploying robot fleets. It emphasizes the importance of safety-critical design, fault-tolerant control loops, and rigorous security measures like secure boot and runtime isolation to protect against physical and digital threats. The later chapters bridge the gap between development and deployment through "Embedded MLOps," detailing how to manage data, version models, and execute over-the-air updates. By utilizing simulation, digital twins, and system identification, the text provides methodologies for overcoming the "sim-to-reality gap."

Ultimately, the work concludes with a series of integrated case studies across mobile, aerial, and industrial robotics. These reference designs illustrate how to harmonize heterogeneous computing platforms—such as MCUs for low-level reflexes and SoCs for high-level intelligence—with real-time middleware like ROS 2 and TSN. The book serves as a practical manual for transforming intelligent intention into reliable, real-time action, ensuring that autonomous agents can operate safely and efficiently in unpredictable physical environments.

What You'll Find Inside:

• The perception-to-action stack as the foundational architecture for transforming sensory data into purposeful physical movement in autonomous agents
• Real-time systems fundamentals including deadlines, jitter, determinism, and scheduling principles critical for reliable edge operation
• Sensor fusion and state estimation techniques ranging from Bayesian filters to deep learning approaches for robust environmental understanding
• Control strategies from PID controllers to Model Predictive Control for precise and safe actuation under constraints
• Hardware considerations for embedded AI including platform selection, model compression, and heterogeneous computing trade-offs

Who's It For:

This book is designed for robotics engineers, embedded systems developers, and autonomous systems designers who need to integrate AI agents with hardware for real-time control and autonomy. It will benefit professionals working on edge AI applications who require both theoretical foundations and practical implementation guidance for building reliable robotic systems operating under strict timing, power, and safety constraints.