Learning Path

This glossary covers a lot of ground. If you’re starting from scratch, follow the stages below in order — each one builds on the last.

If you already know some robotics, jump to the stage where things get unfamiliar.

Stage 1 — What Is a Robot?

Question this stage answers: What are the physical components of a robot arm, and how do we describe its position in space?

Joints and Links The physical building blocks — revolute joints, prismatic joints, and the chains they form

Degrees of Freedom How many independent ways a robot can move

Kinematics How joint angles determine where the end effector ends up in space

End Effector The gripper or tool at the tip of the arm that does the actual work

Coordinate Frames How we describe position and orientation — world frame, base frame, tool frame

Question this stage answers: How does a robot sense its environment — cameras, depth sensors, and building maps?

Computer Vision Feature detection, depth estimation, and how cameras become sensors

Point Cloud 3D data from depth cameras and LiDAR — the primary format for spatial perception

Sensor Fusion Combining camera, IMU, and LiDAR data for robust perception

SLAM Simultaneous Localization and Mapping — building a map while tracking position

Question this stage answers: Given a goal position, how does the robot plan a path and execute it?

Motion Planning Finding collision-free paths from start to goal — RRT, PRM, cuMotion

PID Control The feedback loop that executes planned trajectories precisely

Question this stage answers: Instead of programming every motion, how do we teach a robot by showing it?

This is the paradigm shift in modern robotics. Skip here if you’re interested in imitation learning and AI-trained policies.

Imitation Learning The core paradigm — training from human demonstrations instead of writing rules

Teleoperation How you physically collect demonstrations using leader/follower arms

Robot Training Dataset What episodes are, why quality matters, and how many you need

Action Chunking Predicting sequences of actions — the key insight behind ACT

VLA Models Large vision-language-action models — the frontier of robot learning

Question this stage answers: Once deployed, how does a robot fleet get better over time?

Sim-to-Real Transfer Validate in simulation before risking hardware — the safety gate before deploy

Model Checkpoint What you deploy, how to pick the right one, and how rollback works

OTA Model Update Pushing new weights to a fleet of robots without touching the hardware

Once you understand the concepts, the Guides section shows how these stages play out in practice — with real failures, real data, and real lessons.

Teaching a Robot to Pick Up a Candy Bar Stages 1–5 in one project: 56 demos, 3 training runs, deployed to a real arm

3 Numbers That Tell You How to Optimize Before you try quantization — the latency floor model for AI inference