AI Robot Brain Glossary
A consolidated reference of key terms used throughout Chapter 3. Refer back to this glossary whenever you encounter unfamiliar terminology.
Perception & Sensing
Perception
The process of transforming raw sensor data into meaningful information about the environment. Includes preprocessing, feature extraction, and semantic understanding.
Perception Pipeline
A series of processing stages that convert raw sensor data into actionable information: Sensing → Preprocessing → Feature Extraction → Interpretation.
cv_bridge
A ROS 2 package that converts between ROS image messages and OpenCV image formats, enabling image processing in ROS 2 nodes.
Depth Image
An image where each pixel represents the distance from the camera to that point in the scene. Used for 3D perception and obstacle detection.
Point Cloud
A collection of 3D points representing surfaces in the environment. Generated from LIDAR or depth cameras.
RGB-D Camera
A camera that captures both color (RGB) and depth (D) information simultaneously. Examples: Intel RealSense, Microsoft Kinect.
Feature Extraction
The process of identifying meaningful patterns or characteristics in sensor data, such as edges, corners, or objects.
SLAM (Simultaneous Localization and Mapping)
SLAM
Simultaneous Localization and Mapping - the process of building a map of an unknown environment while simultaneously determining the robot's position within that map.
SLAM Toolbox
A ROS 2 package that implements SLAM algorithms for 2D mapping using laser scan data.
Occupancy Grid
A 2D grid representation of the environment where each cell indicates whether that space is occupied, free, or unknown.
Localization
The process of determining a robot's position and orientation (pose) within a known map.
Mapping
The process of building a representation of the environment from sensor data.
Loop Closure
The process of recognizing when a robot has returned to a previously visited location, enabling correction of accumulated drift in the map.
Pose
The position (x, y, z) and orientation (roll, pitch, yaw) of a robot in space.
Navigation
Nav2
The Navigation 2 stack for ROS 2 - a complete autonomous navigation system including path planning, obstacle avoidance, and recovery behaviors.
Global Planner
A path planning algorithm that computes a complete path from the robot's current position to the goal, considering the known map.
Local Planner
A path planning algorithm that generates immediate motion commands based on the global plan and local sensor data, enabling dynamic obstacle avoidance.
Costmap
A 2D grid that assigns a cost value to each cell, representing how difficult or dangerous it is for the robot to traverse that space.
Static Layer
A costmap layer derived from the pre-built map, representing permanent obstacles.
Obstacle Layer
A costmap layer that marks obstacles detected by sensors in real-time.
Inflation Layer
A costmap layer that adds safety margins around obstacles by inflating their cost values.
Recovery Behavior
Predefined actions the robot takes when navigation fails, such as rotating in place or backing up.
Behavior Tree
A hierarchical structure for organizing robot behaviors and decision-making logic. Used in Nav2 for navigation control.
DWA (Dynamic Window Approach)
A local planning algorithm that generates velocity commands by simulating trajectories and selecting the best one based on a cost function.
Coordinate Frames & Transforms
TF2
The Transform library in ROS 2 that tracks coordinate frame relationships over time.
base_link
The robot's base coordinate frame, typically at the center of the robot's footprint.
map
The fixed world coordinate frame representing the global map.
odom
The odometry frame that tracks the robot's motion from a starting position. Subject to drift over time.
camera_frame
The coordinate frame attached to a camera sensor.
Transform
A mathematical representation of the position and orientation relationship between two coordinate frames.
Reinforcement Learning
Reinforcement Learning (RL)
A machine learning paradigm where an agent learns to make decisions by interacting with an environment and receiving rewards or penalties.
Agent
The learning entity in RL that observes states, takes actions, and receives rewards.
Environment
The world or simulation in which the RL agent operates and learns.
State
A representation of the current situation or configuration of the environment.
Action
A decision or command that the agent can execute to change the environment state.
Reward
A scalar feedback signal that indicates how good or bad an action was in a given state.
Policy
A mapping from states to actions that defines the agent's behavior. Can be deterministic or stochastic.
MDP (Markov Decision Process)
A mathematical framework for modeling decision-making where outcomes depend only on the current state and action.
PPO (Proximal Policy Optimization)
A popular RL algorithm that learns policies by optimizing a clipped objective function. Known for stability and sample efficiency.
SAC (Soft Actor-Critic)
An off-policy RL algorithm that learns both a policy and a value function, optimizing for maximum entropy.
Episode
A complete sequence of states, actions, and rewards from an initial state to a terminal state.
Exploration vs. Exploitation
The trade-off between trying new actions to discover better strategies (exploration) and using known good actions (exploitation).
Simulation & Deployment
Sim-to-Real Transfer
The process of transferring policies or behaviors learned in simulation to real-world robots.
Domain Randomization
A technique for improving sim-to-real transfer by training on varied simulation parameters (lighting, textures, physics).
Reality Gap
The difference between simulated and real-world environments that can cause learned behaviors to fail when deployed.
ONNX (Open Neural Network Exchange)
A format for representing trained neural network models that can be loaded and executed across different frameworks.
Policy Network
A neural network that implements a policy, mapping observations to actions.
Inference
The process of using a trained model to make predictions or decisions on new data.
Algorithms & Techniques
Kalman Filter
An algorithm that estimates the state of a system from noisy measurements, commonly used for sensor fusion and localization.
Particle Filter
A probabilistic localization algorithm that represents the robot's pose as a set of weighted particles.
A* (A-star)
A graph search algorithm used for path planning that finds the shortest path by using heuristics.
Dijkstra's Algorithm
A graph search algorithm that finds the shortest path between nodes, used as a basis for many path planners.
Gradient Descent
An optimization algorithm used to train neural networks by iteratively adjusting parameters to minimize a loss function.
Common Abbreviations
| Abbreviation | Full Name |
|---|---|
| SLAM | Simultaneous Localization and Mapping |
| Nav2 | Navigation 2 |
| RL | Reinforcement Learning |
| PPO | Proximal Policy Optimization |
| SAC | Soft Actor-Critic |
| MDP | Markov Decision Process |
| DWA | Dynamic Window Approach |
| RGB-D | Red Green Blue - Depth |
| LIDAR | Light Detection and Ranging |
| TF | Transform (TF2 = Transform 2) |
| ONNX | Open Neural Network Exchange |
| IMU | Inertial Measurement Unit |