Glossary
Physical AI & Robotics Terminology
This glossary contains 70+ essential terms covering ROS 2, digital twin simulation, NVIDIA Isaac Sim, Vision-Language-Action models, and general robotics concepts. Use the search box below to find terms instantly with fuzzy matching.
Features:
- 🔍 Instant Search - Find terms as you type
- 🔤 Alphabetical Filtering - Browse by first letter
- 📚 Cross-References - Related chapters linked
- 🌐 External Resources - Learn more with curated links
Action
A ROS 2 communication pattern for long-running tasks that provides feedback during execution and allows cancellation. Actions combine request-response (like services) with ongoing feedback updates.
- Module 1: Actions
- Multi-Node Systems
Actuator
A mechanical device that converts electrical signals into physical motion. Common types include DC motors, servo motors, and stepper motors used to control robot joints and movements.
- Module 2: Robot Modeling
Adaptive Monte Carlo Localization
A probabilistic localization algorithm that uses particle filters to estimate a robot's position within a known map. It adapts the number of particles based on localization confidence.
- Module 2: Navigation
AGI
A hypothetical form of artificial intelligence that possesses the ability to understand, learn, and apply knowledge across a wide range of tasks at a human-like level.
- Module 4: VLA Models
Bag File
A file format used in ROS to record and playback topic data. Useful for debugging, testing, and creating reproducible scenarios without re-running live systems.
- Module 1: ROS 2 Fundamentals
Base Link
The primary coordinate frame attached to the robot's base, serving as the reference point for all other frames in the robot's transformation tree (TF tree).
- Module 1: ROS 2 Fundamentals
- Module 2: URDF Modeling
BERT
A transformer-based language model developed by Google that uses bidirectional training to understand context from both left and right of a word in sentences.
- Module 4: VLA Models
Camera Intrinsics
Parameters describing a camera's internal characteristics including focal length, optical center, and lens distortion. Essential for 3D reconstruction and computer vision tasks.
- Module 3: Perception Pipeline
Capstone Project
The culminating project integrating all four course modules (ROS 2, Digital Twin, Isaac Sim, VLA) into a complete autonomous humanoid system capable of natural language control.
- Week 13: Capstone Integration
Cartesian Space
The 3D workspace defined by X, Y, Z coordinates and orientation (roll, pitch, yaw) where the robot end-effector operates, as opposed to joint space.
- Module 2: Manipulation
CLIP
An OpenAI model that learns visual concepts from natural language supervision, enabling zero-shot image classification and vision-language understanding.
- Module 4: VLA Models
Colcon
The build system used in ROS 2 for compiling packages. It supports multiple build types (CMake, Python setuptools, ament) and parallel builds for faster compilation.
- Module 1: ROS 2 Fundamentals
Collision Detection
The process of determining whether objects in a simulated or real environment intersect or come into contact. Critical for motion planning and safety.
- Module 2: Gazebo Simulation
- Module 3: Isaac Sim
Costmap
A 2D grid representation of the environment where each cell has a cost value indicating traversability. Used by navigation algorithms to plan collision-free paths.
- Module 2: Navigation
CUDA
NVIDIA's parallel computing platform and API that allows developers to use GPUs for general-purpose processing, essential for deep learning and physics simulation.
- Setup: Hardware Requirements
- Module 3: Isaac Sim
DDS
The middleware protocol used by ROS 2 for inter-process communication. Provides real-time publish-subscribe messaging with quality-of-service guarantees.
- Module 1: ROS 2 Architecture
Degrees of Freedom
The number of independent parameters that define a robot's configuration. A 6-DOF manipulator can position and orient its end-effector freely in 3D space.
- Module 2: Kinematics
Digital Twin
A virtual replica of a physical system that mimics its behavior, appearance, and dynamics. Used for testing, training, and validation before deploying to real hardware.
- Module 2: Digital Twin Simulation
Direct Kinematics
The process of calculating the end-effector position and orientation given joint angles. The opposite of inverse kinematics.
- Module 2: Kinematics
Domain Randomization
A technique for training models in simulation with randomized parameters (lighting, textures, physics) to improve transferability to real-world environments.
- Module 2: Unity ML-Agents
Embodied AI
AI systems that interact with the physical world through sensors and actuators, as opposed to purely virtual AI. Includes robots, autonomous vehicles, and smart devices.
- Introduction
End Effector
The device at the end of a robotic arm that interacts with the environment, such as a gripper, suction cup, or welding torch.
- Module 2: Manipulation
Extended Kalman Filter
A nonlinear version of the Kalman filter used for state estimation by linearizing around the current estimate. Commonly used for sensor fusion and robot localization.
- Module 3: Sensor Fusion
Fuse.js
A JavaScript library for lightweight fuzzy-search functionality, used in this textbook for the glossary search component with approximate string matching.
- Setup: Glossary Search
Gazebo
An open-source 3D robot simulator that provides physics simulation, sensor models, and integration with ROS. Used for testing algorithms before hardware deployment.
- Module 2: Digital Twin Simulation
Grasp Planning
The process of determining optimal gripper poses and trajectories to securely grasp objects. Involves force analysis, contact modeling, and stability evaluation.
- Module 3: Manipulation
Humanoid Robot
A robot with a body shape resembling the human form, typically including a torso, head, two arms, and two legs. Examples include Atlas, Optimus, and ASIMO.
- Module 4: Humanoid Integration
IMU
A sensor that measures linear acceleration and angular velocity using accelerometers and gyroscopes. Essential for robot orientation and motion estimation.
- Module 1: Sensors
- Module 3: Sensor Fusion
Inverse Kinematics
The process of calculating joint angles required to achieve a desired end-effector position and orientation. Often has multiple solutions or no solution.
- Module 2: Kinematics
- Module 3: Manipulation
Isaac Sim
NVIDIA's GPU-accelerated robotics simulator built on Omniverse, providing photorealistic rendering, accurate physics, and synthetic data generation for perception training.
- Module 3: NVIDIA Isaac Sim
Jetson
NVIDIA's line of embedded computing boards designed for edge AI applications. The Jetson Orin Nano is recommended for this course's edge deployment option.
- Setup: Jetson Orin Nano
Joint Space
The space defined by robot joint angles (q1, q2, ... qn), as opposed to Cartesian space. Motion planning often occurs in joint space.
- Module 2: Kinematics
Kalman Filter
An optimal state estimation algorithm that combines predictions from a model with noisy sensor measurements to estimate system state with minimal error variance.
- Module 3: Sensor Fusion
Lidar
A sensor that measures distance by illuminating targets with laser light and measuring reflections. Produces 2D or 3D point clouds for mapping and obstacle detection.
- Module 3: Perception Pipeline
LLM
A deep learning model trained on vast text corpora to understand and generate human language. Examples include GPT-4, Claude, and PaLM used in VLA architectures.
- Module 4: VLA Models
Localization
The process of determining a robot's position and orientation within a known map or environment using sensor data and probabilistic algorithms.
- Module 2: Navigation
- Module 3: Sensor Fusion
Message
A data structure in ROS that defines the format of information exchanged between nodes via topics. Defined in .msg files with primitive and composite types.
- Module 1: Nodes and Topics
Motion Planning
The process of finding a collision-free path for a robot to move from a start to goal configuration. Algorithms include RRT, PRM, and A*.
- Module 2: Navigation
- Module 3: Manipulation
MoveIt
An open-source motion planning framework for ROS that provides inverse kinematics, collision checking, and manipulation primitives for robot arms.
- Module 2: Manipulation
Navigation2
The official ROS 2 navigation stack providing mobile robot navigation capabilities including localization, path planning, and obstacle avoidance.
- Module 2: Navigation
Node
An executable process in ROS that performs computation. Nodes communicate via topics, services, and actions to form distributed robotic systems.
- Module 1: ROS 2 Fundamentals
Occupancy Grid
A 2D probabilistic map representation where each cell indicates the probability that the space is occupied by an obstacle. Used for navigation and SLAM.
- Module 2: Mapping
Odometry
The estimation of a robot's change in position over time using sensor data such as wheel encoders, IMU, or visual features. Accumulates error over distance.
- Module 1: Sensors
- Module 2: Localization
Omniverse
NVIDIA's platform for 3D simulation and collaboration built on Universal Scene Description (USD). Isaac Sim is built on top of Omniverse.
- Module 3: Isaac Sim Setup
Particle Filter
A probabilistic localization algorithm that represents the robot's pose distribution with a set of weighted particles. Used in AMCL for localization.
- Module 2: Localization
PhysX
NVIDIA's real-time physics engine used in Isaac Sim for accurate rigid body dynamics, collision detection, and contact simulation with GPU acceleration.
- Module 3: Isaac Sim
Point Cloud
A collection of 3D points representing object surfaces or environments, typically captured by lidar or depth cameras. Used for object detection and mapping.
- Module 3: Perception Pipeline
Pose
The combination of a 3D position (x, y, z) and 3D orientation (roll, pitch, yaw or quaternion) that fully describes an object's location in space.
- Module 1: TF2
- Module 3: Pose Estimation
Publish-Subscribe
A messaging pattern where publishers send data to topics without knowing subscribers, and subscribers receive data without knowing publishers. Core ROS 2 pattern.
- Module 1: Nodes and Topics
QoS
Policies in ROS 2 DDS that control message reliability, durability, and history. Different QoS profiles optimize for real-time vs reliable communication.
- Module 1: Advanced Topics
Quaternion
A mathematical representation of 3D orientation using four numbers (w, x, y, z) that avoids gimbal lock. Preferred over Euler angles in robotics.
- Module 1: TF2
- Module 2: Kinematics
Ray Tracing
A rendering technique that simulates light propagation for photorealistic images. Isaac Sim uses GPU ray tracing for sensor simulation and synthetic data.
- Module 3: Isaac Sim
RGB-D Camera
A camera that captures both color (RGB) and depth information for each pixel, enabling 3D reconstruction. Examples include Intel RealSense and Azure Kinect.
- Module 3: Perception
Robot Operating System
An open-source framework for robot software development providing communication infrastructure, hardware abstraction, and package management. ROS 2 is the current version.
- Module 1: ROS 2 Fundamentals
RRT
A sampling-based motion planning algorithm that builds a tree of random configurations to find paths in high-dimensional spaces. Fast but produces jagged paths.
- Module 2: Path Planning
RViz
A 3D visualization tool for ROS that displays sensor data, robot models, transforms, and planned paths. Essential for debugging and monitoring robots.
- Module 1: ROS 2 Tools
Sensor Fusion
The process of combining data from multiple sensors (IMU, GPS, lidar, camera) to produce more accurate and reliable state estimates than any single sensor.
- Module 3: Sensor Fusion
Service
A ROS 2 communication pattern for synchronous request-response interactions. A client sends a request and blocks until receiving a response from the server.
- Module 1: Services
SLAM
The problem of building a map of an unknown environment while simultaneously localizing the robot within that map. Fundamental problem in mobile robotics.
- Module 2: Mapping and Localization
Synthetic Data
Artificially generated training data from simulation rather than real-world collection. Isaac Sim excels at generating labeled synthetic images for perception training.
- Module 3: Isaac Sim
- Module 4: VLA Training
TF2
ROS 2's coordinate frame transformation system that manages the relationships between different reference frames in a robot system over time.
- Module 1: ROS 2 Fundamentals
Topic
A named bus over which nodes exchange messages in a publish-subscribe pattern. Topics enable asynchronous, many-to-many communication in ROS 2.
- Module 1: Nodes and Topics
Transformer
A deep learning architecture using self-attention mechanisms, foundational to modern language models (GPT, BERT) and increasingly used in vision and robotics.
- Module 4: VLA Architecture
Unity ML-Agents
An open-source toolkit for training intelligent agents in Unity environments using reinforcement learning, imitation learning, and neuroevolution.
- Module 2: Digital Twin Simulation
URDF
An XML format for describing robot models including links, joints, visual meshes, collision geometry, and inertial properties. Standard for ROS robots.
- Module 2: Robot Modeling
Vision-Language-Action
A class of models that integrate vision, language understanding, and action prediction to enable robots to follow natural language commands for manipulation tasks.
- Module 4: VLA Models
Visual Servoing
A control technique that uses visual feedback from cameras to guide robot motion, enabling tasks like object tracking and precision manipulation.
- Module 3: Vision-Based Control
VRAM
Memory dedicated to the graphics card for storing textures, framebuffers, and computation data. Isaac Sim requires 12GB+ VRAM for high-fidelity simulation.
- Setup: Hardware Requirements
Workspace
A directory structure containing ROS 2 packages that are built together. Consists of src/ (source code), build/ (build artifacts), install/ (binaries), and log/ folders.
- Module 1: ROS 2 Setup
YOLO
A family of real-time object detection models that process images in a single forward pass. YOLOv8 is commonly used for robotic perception tasks.
- Module 3: Object Detection
Zero-Shot Learning
The ability of a model to recognize or perform tasks it wasn't explicitly trained on by leveraging semantic knowledge. VLA models exhibit zero-shot manipulation capabilities.
- Module 4: VLA Models