Skip to main content

Glossary

This glossary provides definitions for key terms used throughout the Physical AI & Humanoid Robotics book.

A​

Action Space​

The set of all possible actions that an agent can take in an environment. In robotics, this typically refers to the space of possible motor commands or movement primitives.

Adaptive Planning​

A planning approach that adjusts plans based on environmental changes, unexpected obstacles, or new information during execution.

C​

Cognitive Planning​

High-level reasoning process that determines sequences of actions to achieve goals, often involving abstract reasoning and long-term decision making.

Computer Vision​

Field of study focused on enabling computers to interpret and understand visual information from the world, often using cameras and image processing algorithms.

D​

Deep Learning​

Subset of machine learning that uses neural networks with multiple layers to learn complex patterns from data.

Digital Twin​

Virtual replica of a physical system that simulates its behavior and characteristics for analysis, testing, and optimization.

Domain Randomization​

Technique used in simulation to randomize various aspects of the environment to improve transfer learning from simulation to reality.

E​

Embodied Intelligence​

Intelligence that emerges from the interaction between an agent's cognitive processes, its physical form, and its environment.

G​

Gazebo​

Open-source 3D robotics simulator that provides realistic physics simulation and rendering capabilities.

I​

Isaac Sim​

NVIDIA's simulation environment for robotics, autonomous vehicles, and AI applications featuring high-fidelity physics and rendering.

Isaac ROS​

Collection of packages and tools that facilitate the integration of NVIDIA Isaac technologies with the Robot Operating System (ROS).

L​

Language Understanding​

Component of AI systems that processes and interprets human language to extract meaning and intent.

Latent Space​

Low-dimensional representation of high-dimensional data, often used in machine learning models to capture essential features.

LLM (Large Language Model)​

Advanced neural network models trained on vast amounts of text data to understand and generate human-like language.

M​

Multi-Modal Learning​

Learning approach that combines information from multiple sensory modalities (e.g., vision, audio, tactile) to improve performance.

N​

Collection of algorithms and components responsible for path planning, obstacle avoidance, and motion control for mobile robots.

P​

Perception Pipeline​

Series of processing steps that transform raw sensor data into meaningful information about the environment.

Physical AI​

Application of artificial intelligence techniques to physical systems, particularly robots that interact with the real world.

Point Cloud​

Set of data points in space, typically representing the external surface of an object, commonly used in 3D sensing and mapping.

R​

Reinforcement Learning​

Machine learning paradigm where an agent learns to make decisions by interacting with an environment and receiving rewards.

Robot Operating System (ROS)​

Flexible framework for writing robot software that provides services for hardware abstraction, device drivers, and message passing.

ROS 2​

Second generation of the Robot Operating System with improved real-time capabilities, security, and multi-robot support.

S​

Semantic Segmentation​

Computer vision task that assigns a label to each pixel in an image, identifying which object class each pixel belongs to.

Sim-to-Real Transfer​

Process of transferring skills or behaviors learned in simulation to real-world robotic systems.

SLAM (Simultaneous Localization and Mapping)​

Process by which a robot builds a map of an unknown environment while simultaneously keeping track of its location within that map.

T​

Transformer Architecture​

Deep learning architecture that uses attention mechanisms to process sequential data, widely used in natural language processing.

Trajectory Planning​

Process of determining a sequence of positions, velocities, and accelerations for a robot to follow over time.

V​

Vision-Language-Action (VLA)​

System architecture that integrates visual perception, language understanding, and action execution in a unified framework.

Visual Servoing​

Control technique that uses visual feedback to control the motion of a robot, often for tasks requiring precise positioning.

W​

Whisper​

OpenAI's automatic speech recognition system designed for robust speech-to-text conversion.