- Actuator
- The part that makes a robot move, such as a motor.
- Algorithm
- A step-by-step recipe a computer follows to solve a problem.
- Autonomous
- Able to do a task without a person steering every move.
- Classification
- Putting something into a category, such as "ripe" or "unripe".
- Computer vision
- Methods that let a computer understand pictures and video.
- Dataset
- The collection of examples used to teach a model.
- Deep learning
- Machine learning that uses many layers of simple units, called a neural network.
- Edge AI
- Running AI on the device itself instead of in the cloud.
- End effector
- The "hand" at the end of a robot arm, such as a gripper.
- Imitation learning
- Learning a skill by copying demonstrations from a person.
- Inference
- Using a trained model to make a prediction. This is the model "working".
- Kalman filter
- A maths method that blends noisy sensor readings into one better estimate.
- Lidar
- A sensor that measures distance by timing how long laser light takes to bounce back.
- LLM (large language model)
- The kind of AI behind chat assistants. Good at language and planning.
- Localisation
- Working out where the robot is.
- Machine learning
- Computers finding patterns from examples instead of following hand-written rules.
- Model
- The learned "recipe" produced by training. You feed it input and it gives an output.
- Neural network
- A model made of many connected simple units whose strengths are learned from data.
- Object detection
- Finding and naming things in an image, usually with a box around each.
- Path planning
- Finding a safe route from one place to another.
- Perception
- Everything a robot does to understand its surroundings from sensors.
- Quantisation
- Shrinking a model by using fewer bits per number so it runs faster on small chips.
- Reinforcement learning
- Learning by trial, error and a reward score.
- Reward
- The score that tells a learning robot whether an action was good.
- ROS 2
- Free software that helps different parts of a robot talk to each other.
- Sensor fusion
- Combining several sensors so their errors cancel out.
- Sim-to-real
- Moving a skill learned in simulation onto a real robot.
- SLAM
- Building a map and finding your place on it at the same time.
- Training
- The process of letting a model learn from examples.
- Training data
- The examples a model learns from. Poor data gives a poor model.