Strategic Objectives
• Bridge the gap between raw geometric data and high-level conceptual reasoning.
• Implement state-of-the-art deep learning architectures for object recognition.
• Master the integration of spatial geometry with semantic label fusion.
• Develop autonomous systems that navigate with human-like environmental context.
The Core Challenge
Robots can move through space, but they often fail to comprehend what they see, leading to fragile and context-blind AI systems.
The Evolution of Perception
The Geometric Foundation of Robotic Awareness
This section traces the origins of robotic perception through the emergence of spatial representation techniques that allowed machines to transform sensor measurements into usable maps. It examines the shift from raw observations to structured geometric models, including occupancy representations, localization, and the challenge of building reliable spatial knowledge in uncertain environments. The focus is on how early robotic systems answered the question of 'where am I?' while remaining largely blind to the meaning of the spaces they navigated.
The Limits of a World Made Only of Coordinates
This section explores the conceptual limitations of purely geometric maps and explains why spatial precision does not equal intelligence. It investigates the gap between knowing an object's position and understanding its role, identity, or function within a scene. By examining the demands of real-world autonomy, the discussion reveals why robots require richer representations that connect physical structure with contextual meaning, enabling them to reason about rooms, objects, obstacles, and interactions rather than simply measure distances.
The Emergence of Semantic Spatial Intelligence
This section introduces the transition from metric mapping to semantic understanding as a fundamental evolution in robotics. It examines how modern systems combine mapping, computer vision, object recognition, and artificial intelligence to create representations where places and objects carry conceptual meaning. The chapter concludes by establishing semantic intelligence as the bridge between perception and action, showing why future robots must not only navigate environments but comprehend them as organized worlds of entities, relationships, and purposes.
Foundations of Computer Vision
From Photons to Pixels: The Birth of Machine Vision
This section introduces the fundamental pipeline that allows machines to perceive visual reality. It explores the transition from light captured by sensors into numerical image data, explaining the role of cameras, image formation, sampling, resolution, color representation, and the limitations imposed when converting continuous environments into discrete computational structures. The focus is on developing intuition for how visual information enters an artificial system before higher-level intelligence can interpret it.
Transforming Visual Signals into Meaningful Patterns
This section examines the core image processing operations that prepare visual data for intelligent interpretation. It covers enhancement, filtering, feature extraction, segmentation, and representation techniques that convert raw pixels into structured patterns. The discussion emphasizes how these processes create the intermediate layers of understanding required for machines to identify objects, boundaries, textures, and spatial relationships within complex environments.
Building the Visual Intelligence Layer
This section connects foundational computer vision methods to the broader goals of semantic robotics. It explores how processed visual information becomes the basis for object recognition, scene interpretation, and spatial reasoning. The chapter concludes by showing how traditional vision pipelines have evolved alongside machine learning and deep neural networks, creating systems capable of transforming visual observations into actionable knowledge for autonomous machines.
Deep Learning Architectures
From Pixels to Meaning: The Rise of Hierarchical Feature Learning
This section introduces the fundamental shift from handcrafted visual features to learned representations that allow robots to interpret complex environments. It explores how layered neural architectures progressively extract edges, patterns, objects, and higher-level semantic structures from camera and sensor inputs, establishing the foundation for scene understanding and semantic mapping.
Architectures Behind Robotic Perception
This section examines the major architectural principles that make modern recognition systems effective, including convolutional structures, pooling strategies, training mechanisms, and specialized network designs. It connects these concepts to robotic applications by explaining how different architectures influence object detection, scene classification, segmentation, and the extraction of spatially meaningful information for autonomous systems.
Engineering Intelligence: Optimizing Deep Models for Semantic Mapping
This section focuses on the practical challenges of deploying deep learning architectures in intelligent robots. It explores model optimization, computational constraints, transfer of learned knowledge, and the integration of recognition networks into larger perception pipelines. The discussion emphasizes how architectural choices determine a robot’s ability to build reliable semantic maps and operate effectively in dynamic environments.
Semantic Segmentation
From Pixels to Perception: Teaching Robots to Understand Visual Boundaries
This section introduces semantic segmentation as a foundational capability for robotic scene understanding, explaining how machines move beyond recognizing entire images or isolated objects to assigning meaning at every pixel. It explores the transition from traditional image partitioning methods to modern AI-driven approaches that allow robots to distinguish surfaces, objects, and environmental structures with fine-grained precision. The discussion establishes why pixel-level interpretation is essential for navigation, manipulation, and interaction in complex environments.
Mapping Meaning onto Every Pixel: The Architecture of Semantic Segmentation Systems
This section examines the core computational principles behind semantic segmentation systems, including how neural networks analyze visual patterns and produce dense classifications across an image. It explores the relationship between image features, object categories, and boundary accuracy, highlighting the role of convolutional architectures, encoder-decoder designs, and learning-based representations in creating detailed semantic maps. The section connects algorithmic processes with robotic applications where precise identification of objects such as doors, furniture, and obstacles determines effective decision-making.
Building Spatial Intelligence Through Semantic Understanding
This section explores how semantic segmentation transforms visual perception into actionable knowledge for autonomous robots. It explains how pixel-level world models support tasks such as navigation, scene reconstruction, object interaction, and safe operation in dynamic environments. The discussion focuses on the practical challenges of achieving reliable segmentation in real-world conditions, including variations in lighting, complex backgrounds, and ambiguous boundaries, while showing how improved semantic understanding advances the broader goal of intelligent robotic autonomy.
Spatial Geometry and Transform
The Mathematical Language of Physical Space
This section establishes how robots represent the structure of the world through mathematical models of space. It explores points, vectors, coordinate systems, distances, angles, and spatial relationships as the fundamental vocabulary that allows an intelligent machine to convert visual observations into measurable environmental knowledge. The discussion connects classical geometric principles with modern scene understanding, showing how abstract mathematics becomes the foundation for perception, localization, and interaction.
Transforming Perspectives Across the Three-Dimensional World
This section examines the critical role of spatial transformations in robotics and AI perception. It explains how robots translate and rotate objects, sensors, and environments between different coordinate frames, enabling consistent interpretation of scenes from changing viewpoints. The section focuses on transformations as the bridge between raw sensor data and stable semantic understanding, including their importance in mapping, navigation, object tracking, and autonomous decision-making.
From Geometric Models to Semantic Spatial Intelligence
This section explores how geometric reasoning enables robots to attach meaning to the world they perceive. It investigates the relationship between spatial geometry and semantic interpretation, showing how accurate 3D positioning allows AI systems to understand object locations, environmental structures, and actionable relationships. The chapter concludes by connecting geometric precision with higher-level capabilities such as navigation, manipulation, and embodied intelligence.
Simultaneous Localization and Mapping
The Paradox of Knowing Where You Are While Discovering Where You Are Going
This section introduces SLAM as a fundamental reasoning problem for autonomous robots: the need to estimate position while constructing an unknown environment model. It explores why navigation requires a continuous feedback loop between perception, motion estimation, and spatial representation, establishing SLAM as the geometric foundation that allows robots to transform raw sensor data into an understanding of place.
Building a Living Map from Motion and Measurement
This section examines the computational architecture behind SLAM systems, including how robots combine camera, lidar, inertial, and other sensor information to estimate trajectories and create maps. It explains the evolution from geometric mapping toward richer representations where objects, surfaces, and semantic meanings become integrated into the spatial model, enabling robots to move beyond coordinates toward contextual understanding.
When Maps Become Meaningful: Semantic SLAM and Intelligent Exploration
This section explores the future direction of SLAM as a bridge between spatial geometry and semantic intelligence. It analyzes how modern robots use learned representations, object recognition, and scene understanding to create maps that describe not only where things are but what they are and how they relate. The section highlights how semantic SLAM enables robots to reason about environments, adapt to change, and perform complex tasks in human-centered spaces.
Object Recognition and Labeling
From Visual Patterns to Meaningful Entities
This section introduces object recognition as a foundational capability in semantic robotics, explaining how machines move beyond low-level visual features such as edges, colors, and textures to identify distinct entities with practical meaning. It explores the transition from image appearance to object identity, the role of feature extraction, classification, and learned representations, and why recognizing objects is essential for robots that must interpret and interact with real-world environments.
The Architecture of Object Detection and Labeling
This section examines the technical pipeline that enables robots to identify objects within complex environments. It covers the relationship between detecting an object's location and assigning semantic labels, including approaches based on machine learning and deep neural networks. The discussion highlights bounding regions, category prediction, confidence estimation, and the challenges of recognizing objects under changing conditions such as different viewpoints, lighting, scale, and partial visibility.
Recognizing Objects as Functional Elements of Space
This section explores how object recognition becomes meaningful when integrated into complete scene understanding. It explains how identifying chairs, tables, tools, appliances, and other entities allows robots to reason about environments, plan actions, and interact safely with humans. The chapter concludes by examining the importance of semantic labels as building blocks for spatial intelligence, enabling robots to understand not only what objects are but also their roles within a larger context.
Point Cloud Processing
From Sensor Measurements to Spatial Representations
This section introduces point clouds as the fundamental representation of three-dimensional environments captured by LiDAR, RGB-D cameras, and other sensing technologies. It explains how millions of individual depth measurements become spatial structures that robots can interpret, while examining the challenges of noise, sparsity, irregular organization, and the absence of traditional image-grid assumptions. The focus is on why point clouds are a critical bridge between physical reality and semantic scene understanding.
Preparing and Transforming 3D Point Data
This section explores the processing pipeline required to convert raw point clouds into useful representations for artificial intelligence systems. It covers filtering, noise reduction, downsampling, registration, coordinate transformations, and alignment methods that allow multiple observations to be combined into coherent spatial models. The discussion emphasizes how preprocessing decisions influence downstream tasks such as object recognition, mapping, localization, and semantic labeling in autonomous robots.
Teaching Machines to Understand Point-Based Worlds
This section examines how modern AI methods extract meaning from point clouds by learning geometric patterns, identifying objects, and assigning semantic labels to regions of space. It explores the relationship between point cloud processing and deep learning approaches designed for irregular 3D data, highlighting applications in robotics, autonomous navigation, digital mapping, and embodied intelligence. The section concludes by showing how raw spatial points become actionable knowledge for robots operating in complex environments.
Probabilistic Data Fusion
The Uncertainty Problem: Why Robots Must Reason Beyond Individual Sensors
This section introduces the fundamental challenge of robotic perception: every sensor captures only a partial and imperfect view of the world. It explores sources of uncertainty such as measurement noise, environmental changes, occlusion, and sensor limitations, establishing why semantic robots require probabilistic reasoning rather than deterministic interpretations. The discussion frames data fusion as a mechanism for building confidence, resolving ambiguity, and maintaining a coherent understanding of dynamic environments.
The Mathematics of Trust: Probabilistic Fusion Across Multiple Data Streams
This section examines the computational foundations that allow robots to merge diverse observations into a unified representation. It explores probabilistic models, Bayesian reasoning, filtering approaches, and confidence estimation as tools for evaluating competing sensor inputs. The focus is on how algorithms determine which observations to trust, how uncertainty is propagated through perception pipelines, and how fused data improves the reliability of semantic maps used for navigation and interaction.
Building Resilient Semantic Maps Through Intelligent Fusion
This section connects probabilistic data fusion to the broader mission of semantic robotics: creating persistent, accurate, and context-aware models of the environment. It explores how fused sensor data supports scene understanding, object recognition, localization, and autonomous decision-making. The section highlights real-world challenges such as conflicting perceptions, sensor failures, and changing environments while showing how intelligent fusion enables robots to remain dependable under uncertainty.
Graph-Based Representations
From Pixels to Relationships: The Need for Graph Thinking
This section introduces the conceptual shift from traditional image representations based on pixels, grids, and geometric coordinates toward relational models that capture meaning. It explains why robots require graph-based structures to understand that objects are not isolated entities but components of an interconnected environment. The discussion frames graphs as a bridge between perception and reasoning, enabling machines to encode relationships such as proximity, containment, support, ownership, and interaction.
Building Semantic Scene Graphs for Robotic Intelligence
This section explores how robots transform visual observations into semantic graphs where nodes represent recognized objects and edges represent meaningful relationships. It examines how scene graphs combine object recognition, spatial reasoning, and contextual understanding to create machine-readable descriptions of environments. Examples illustrate how a robot can infer that a cup is on a table, a chair is beside a desk, or a person is interacting with an object by reasoning over graph structures rather than isolated detections.
Graph Reasoning as the Foundation of Spatial Intelligence
This section examines how graph-based representations enable higher-level robotic reasoning, including navigation, planning, memory formation, and interaction with dynamic environments. It explores how graphs allow AI systems to infer missing information, update environmental knowledge, and combine perception with symbolic reasoning. The section positions graph representations as a core architecture for semantic robots that must understand not only where objects are, but how they relate and what those relationships imply for action.
Ontologies and Knowledge Bases
Building a Formal Language for Reality
This section introduces ontologies as the semantic foundations that allow robots to move beyond raw sensory data and construct meaningful representations of objects, environments, and relationships. It explores how concepts, categories, properties, and hierarchical structures create a shared vocabulary that enables machines to identify what exists in the world and reason about its significance.
Knowledge Bases as the Memory Architecture of Intelligent Machines
This section examines how robotic knowledge bases store and organize semantic information gathered from perception systems, databases, and learned experiences. It explains how linked concepts, logical assertions, and contextual relationships allow robots to interpret scenes, infer hidden information, and make decisions based on more than immediate sensory input.
From Static Categories to Adaptive World Models
This section explores the future of robotic semantics, where ontologies become dynamic systems capable of adapting to new environments, experiences, and discoveries. It analyzes the challenges of ambiguity, changing contexts, and integrating human knowledge with machine learning to create robots that can continuously refine their understanding of the world.
Visual Odometry
The Eyes That Measure Motion
Introduces visual odometry as the process that allows robots to infer their movement through the environment by analyzing consecutive camera observations. This section establishes the relationship between perception, motion estimation, and semantic mapping, explaining why accurate visual tracking is essential for robots operating without reliable external positioning systems. It explores the transition from raw image streams to meaningful estimates of translation, rotation, and trajectory.
From Pixels to Pose Estimation
Examines the core mechanisms that enable a robot to calculate its changing position from visual data. The section explores feature detection and matching, geometric reconstruction, camera motion estimation, and the use of mathematical models to determine spatial transformation between viewpoints. It explains how these processes convert visual differences into reliable estimates of a robot's pose while addressing challenges such as illumination changes, texture limitations, and dynamic environments.
Maintaining Spatial Truth in a Moving World
Explores how visual odometry becomes a foundational layer for advanced robotic understanding by preserving spatial consistency as robots navigate complex environments. This section connects motion estimation with simultaneous localization and mapping, semantic scene interpretation, and long-term autonomous operation. It analyzes the consequences of drift, the importance of sensor fusion, and how reliable visual tracking enables robots to build, update, and reason about the world around them.
Scene Reconstruction
From Visual Evidence to Spatial Reality
This section introduces the principles behind scene reconstruction as the process of converting images, depth measurements, and sensor observations into coherent 3D representations. It examines how robots combine multiple viewpoints, geometric constraints, and perception pipelines to infer hidden structures and create reliable spatial models. The discussion frames reconstruction as a foundation for semantic robotics, where geometry becomes the bridge between raw perception and intelligent interaction.
Engineering Persistent Digital Twins
This section explores how reconstructed environments evolve into digital twins: persistent, semantically enriched replicas that represent not only shape and location but also objects, relationships, and functional properties. It covers the integration of point clouds, meshes, volumetric models, and semantic labels to produce intelligent spatial databases that robots can query, update, and reason about over time.
Reconstruction as the Foundation of Autonomous Understanding
This section examines the practical role of scene reconstruction in autonomous systems, emphasizing how digital twins support navigation, manipulation, simulation, planning, and human-robot collaboration. It analyzes the challenges of maintaining accurate models in changing environments, including uncertainty, sensor limitations, and real-time updates, while presenting reconstruction as a critical capability for robots that must understand and operate within complex physical worlds.
Real-Time Constraints
The Physics of Instantaneous Intelligence
Explores why real-time semantic robotics depends on more than raw computational power. This section examines the relationship between perception latency, decision cycles, control responsiveness, and the physical demands of operating in dynamic environments. It frames real-time performance as a system-level requirement where delays in sensing, mapping, inference, and action can directly affect safety, reliability, and mission success.
Engineering Semantic Mapping for the Edge
Examines the strategies required to bring advanced scene understanding capabilities from powerful computing platforms to resource-constrained robotic edge devices. This section covers optimization approaches such as efficient perception pipelines, adaptive computation, hardware-aware algorithms, and selective processing techniques that preserve critical semantic information while meeting strict performance budgets.
Balancing Intelligence Depth with Reaction Speed
Investigates how semantic robots can balance sophisticated reasoning with immediate responsiveness. This section presents architectural principles for layered intelligence, prioritizing urgent tasks, managing computational workloads, and dynamically adjusting perception accuracy based on environmental demands. It highlights the future of autonomous systems that continuously negotiate between deeper understanding and faster action.
Instance Segmentation
From Object Recognition to Individual Identity
This section introduces the conceptual leap from detecting object categories to identifying separate physical instances within the same category. It explores why a semantic robot must understand that two identical chairs are not merely two examples of the same label but distinct entities with separate locations, states, and interaction histories. The discussion establishes instance segmentation as a foundation for precise perception, manipulation, and spatial reasoning.
Separating the Indistinguishable in Complex Scenes
This section examines how instance segmentation systems separate visually similar objects through advanced perception pipelines. It explores the role of masks, feature extraction, contextual reasoning, and deep learning architectures in creating distinct representations for objects that share appearance, shape, and category. The focus is on how robots overcome ambiguity when multiple identical items appear together in cluttered environments.
Turning Instance Awareness into Robotic Intelligence
This section connects instance segmentation with real-world robotic capabilities by showing how individual object awareness enables safer and more reliable actions. It explores applications such as warehouse inventory, multi-object grasping, collaborative robotics, and autonomous navigation where confusing one object for another can lead to operational failure. The chapter concludes by positioning instance segmentation as a critical bridge between visual perception and intelligent physical interaction.
Depth Perception and Stereo Vision
The Geometry of Seeing Distance
This section establishes the foundations of depth perception by examining how a flat image can encode three-dimensional structure. It explores the relationship between visual cues, camera geometry, and spatial inference, showing how both biological observers and robotic systems estimate distance, scale, and position from visual information. The discussion connects human perception mechanisms with the computational challenges faced by autonomous machines attempting to reconstruct the physical world.
Stereo Vision as a Computational Sense of Volume
This section explores stereo vision as a core technique for robotic depth estimation, focusing on how multiple viewpoints enable machines to calculate spatial relationships. It examines camera calibration, image correspondence, disparity measurement, and depth reconstruction as essential components of semantic robot perception. The section emphasizes how stereo systems bridge the gap between visual observation and geometric mapping, enabling robots to navigate, localize objects, and understand environmental structure.
Building Reliable Depth Models for Autonomous Mapping
This section examines how depth perception techniques become practical tools for autonomous systems through integration with mapping, localization, and scene understanding pipelines. It addresses the limitations of real-world depth estimation, including uncertainty, environmental conditions, and sensor constraints, while exploring how improved algorithms create more robust geometric representations. The focus is on transforming depth data into actionable spatial intelligence for semantic robots operating in complex environments.
Indoor Scene Understanding
The Hidden Grammar of Indoor Worlds
This section explores indoor environments as structured semantic systems rather than random collections of objects. It examines how architectural layouts, room purposes, furniture arrangements, and human routines create recurring patterns that enable robots to infer meaning from partial visual information. The discussion focuses on how semantic algorithms can leverage contextual relationships—such as object placement, functional zones, and environmental cues—to understand what a space represents and what actions are likely to occur within it.
Semantic Mapping Inside Confined Spaces
This section examines the computational challenges of transforming indoor perception into actionable spatial intelligence. It covers how robots identify rooms, recognize objects, estimate their relationships, and construct semantic maps that connect visual observations with likely meanings. The chapter develops the idea that indoor understanding depends on combining geometry, object recognition, and contextual reasoning to overcome ambiguity caused by occlusion, changing arrangements, and limited viewpoints.
Predictive Intelligence for Adaptive Indoor Navigation
This section explores how semantic robots move beyond recognition toward prediction by learning the probabilistic structure of indoor environments. It analyzes how knowledge of common layouts allows autonomous systems to anticipate object locations, infer missing information, and make safer decisions in unfamiliar spaces. The focus is on future-facing applications where robots use learned environmental patterns to interact naturally with homes, workplaces, healthcare facilities, and other human-designed interiors.
Dynamic Environments
The Moving World Problem
This section introduces the fundamental challenge of maintaining semantic understanding when the physical world continuously changes. It examines how autonomous robots must distinguish between permanent structures, temporary obstacles, moving agents, and rearranged objects while preserving a reliable representation of space. The discussion frames dynamic mapping as a core requirement for robots operating in homes, workplaces, and public environments where uncertainty and change are unavoidable.
Adaptive Maps and Temporal Intelligence
This section explores how robots combine perception, localization, and semantic reasoning to update their understanding of environments over time. It covers approaches for tracking moving objects, revising spatial beliefs, managing outdated information, and integrating temporal context into semantic maps. The focus is on designing representations that allow robots to remember what belongs to the environment while adapting to what changes within it.
Navigating Through Constant Change
This section examines how dynamic scene understanding influences robotic decision-making and motion generation. It explains how robots must plan safe and intelligent actions when people move unpredictably, pathways become blocked, or objects shift positions. The chapter concludes by connecting motion planning with semantic intelligence, showing how future autonomous systems will require not only maps of where things are, but an understanding of how and why environments transform.
Transfer Learning
Beyond Retraining: The Logic of Knowledge Transfer
This section introduces transfer learning as a strategy for overcoming the limitations of training autonomous systems from scratch. It explores how semantic robots can reuse visual features, spatial representations, and learned environmental patterns from previous experiences to accelerate adaptation in new locations. The discussion frames transfer learning as a bridge between laboratory-trained intelligence and scalable real-world deployment.
Crossing Environmental Boundaries with Adaptive Intelligence
This section examines how transfer learning enables semantic mapping systems to function across changing environments, from different buildings and cities to unfamiliar terrain and operating conditions. It explores domain shifts, adaptation strategies, and the challenges of preserving accuracy when sensor data, object appearances, layouts, and environmental contexts differ from the original training setting.
Building Robots That Learn Once and Deploy Everywhere
This section explores the practical impact of transfer learning on autonomous robots operating at scale. It covers how transferred intelligence reduces data requirements, lowers development costs, and enables faster deployment of scene understanding systems in real-world applications. The section also considers future directions where robots continuously refine their spatial intelligence through accumulated experiences.
Evaluation and Benchmarking
Defining Success in Semantic Scene Understanding
Establishes the foundations of evaluating semantic robots by translating abstract capabilities such as object recognition, spatial awareness, and environmental interpretation into measurable performance objectives. This section explores how evaluation frameworks must reflect the real-world demands of autonomous systems, including accuracy, robustness, consistency, and the ability to construct reliable semantic maps.
The Metrics That Reveal Mapping Performance
Examines the mathematical and experimental metrics used to validate scene understanding systems, including classification quality, detection performance, segmentation accuracy, localization precision, and semantic consistency across environments. The section explains how measures such as precision, recall, accuracy, and error analysis can be adapted to evaluate intelligent robots operating in complex and dynamic spaces.
Benchmarking Autonomous Intelligence in the Real World
Explores how benchmarks, datasets, and evaluation protocols transform experimental results into evidence of practical capability. This section focuses on limitations of isolated metrics, the importance of diverse testing environments, failure analysis, and continuous validation methods needed to ensure semantic robots remain dependable when deployed in unpredictable real-world conditions.
The Future of Spatial AI
From Spatial Perception to Machine Understanding
This section examines the transition from robots that merely detect objects and navigate environments to intelligent systems that understand the meaning, purpose, and relationships embedded within physical spaces. It explores how semantic mapping provides machines with contextual models of the world, allowing them to reason about environments, predict outcomes, and interact with humans and objects in increasingly sophisticated ways.
Embodied Intelligence and the Path Toward Autonomous Agents
This section explores how embodiment transforms spatial AI from an abstract computational process into a foundation for intelligent action. It analyzes the role of sensors, interaction, memory, and environmental experience in enabling machines to build internal models of reality. The discussion focuses on how robots with semantic awareness can move beyond task execution toward adaptive behavior, planning, and autonomous decision-making.
The Spatial Frontier of Artificial General Intelligence
This concluding section investigates the possibility that spatial intelligence may represent a critical pathway toward artificial general intelligence. It considers the remaining challenges of creating machines that possess flexible understanding across environments, transfer knowledge between domains, and form robust interpretations of the physical world. The section presents semantic mapping as a bridge between perception, cognition, and the emergence of truly autonomous intelligent systems.
Explore Research Ecosystem
Available eBook Editions
Arabic
English
French
German
Italian
Japanese
Korean
Portuguese
Spanish
