Structured learning paths, essential software tools, recommended video channels, and a complete glossary of robotics terminology — curated for serious learners at every level.
Each path is a curated sequence of 2BRO articles, external readings, and exercises that takes you from where you are to where you want to be. Follow them in order, or jump to what matters most.
No background assumed. You'll finish with a solid mental model of how robots work — mechanically, electronically, and computationally — and be ready to go deeper into any area of the field.
For people with some technical background who want to understand how modern autonomous robots navigate, perceive, and make decisions. Covers SLAM, path planning, and computer vision in depth.
For technically literate readers who want to understand where the field is moving. Covers foundation models for robotics, reinforcement learning at scale, and the convergence of large models with physical systems.
A non-technical path focused on how robots are being deployed in the real world — what's happening in manufacturing, medicine, agriculture and logistics, and what the economic implications are.
A practical path for people considering entering the field. Covers the key technical disciplines, the software ecosystem, and what employers are actually looking for in robotics engineers in 2025.
A focused path for engineers and researchers who want to understand the full technical stack behind humanoid robots — from whole-body control and dexterous manipulation to the AI systems that drive them.
The software stack every robotics engineer should know. Descriptions are honest about what each tool is good at and where it falls short.
The de facto standard middleware for robotics. ROS 2 (Humble, Iron) is the current recommended version. Essential for any serious robotics development — provides hardware abstraction, message passing, simulation integration, and a vast package ecosystem.
The standard simulation environment for ROS-based development. Gazebo Harmonic (now Ignition) provides realistic physics, sensor simulation, and tight ROS integration. Essential for developing and testing algorithms before deploying on hardware.
Python is the dominant scripting language in robotics. OpenCV provides the computer vision primitives. Together they are the entry point for most perception work — image processing, feature detection, and the bridge to deep learning frameworks.
The dominant framework for robotics AI research. More flexible than TensorFlow for research purposes. Used by most major labs for training perception models, policy networks, and generative models applied to robotic control.
The standard motion planning framework for ROS 2. Provides inverse kinematics solvers, path planners (including OMPL), collision checking, and trajectory execution for manipulator arms. The go-to for manipulation research and industrial applications.
ROS 2's navigation framework for mobile robots. Handles localisation (AMCL, SLAM toolbox), path planning (A*, NavFn, Smac), and obstacle avoidance. The standard starting point for any autonomous mobile robot project.
Not every channel that covers robotics is worth watching. These are the ones we return to — selected for accuracy, depth, and the ability to explain difficult concepts without sacrificing substance.
Primary source video of the most advanced locomotion systems currently operating. No hype, no commentary — just demonstrations that show precisely what the state of the art looks like in 2025.
University of Bonn professor. Exceptional lecture series on SLAM, mobile robotics, and probabilistic robotics. Rigorous, accurate, and clearly structured. The best free university-level robotics content available online.
Long-form interviews with leading robotics researchers and engineers. Pieter Abbeel (co-founder of Covariant) asks better technical questions than almost any other interviewer in the space. Essential for understanding where research is heading.
Practical, hands-on ROS 2 content from a working robotics engineer. Excellent for bridging the gap between conceptual understanding and actual implementation. Among the clearest ROS 2 tutorial content currently available.
Russ Tedrake's underactuated robotics lectures. Highly mathematical, covers nonlinear dynamics, trajectory optimisation, and the control theory underlying modern robot locomotion. The clearest rigorous treatment of these topics freely available.
Every significant term in robotics explained clearly and accurately. Jargon-free definitions wherever possible; technical precision where it matters.
A device that converts energy into physical motion. Common types include electric motors (DC, stepper, servo), hydraulic pistons, and pneumatic cylinders. The actuator is the component that actually moves part of a robot in response to control signals.
Operating independently without continuous human input. An autonomous robot makes decisions based on its own sensor data and processing rather than real-time commands. Levels of autonomy range from teleoperation through shared control to full autonomy.
The number of independent axes along which a robot can move. A fully mobile robot in 3D space has 6 DoF (three translational, three rotational). A typical industrial arm has 6 joints and therefore 6 DoF, giving it the ability to position its end effector at any point in its workspace at any orientation.
A category of machine learning that uses neural networks with many layers to learn representations of data. In robotics, deep learning is applied to perception tasks (object detection, scene understanding) and increasingly to control (learning policies directly from sensor input).
The device at the end of a robot arm that interacts with the environment — a gripper, welding torch, suction cup, or tool. End effectors are task-specific and are often swapped depending on the operation required.
The mathematical problem of determining the joint angles required to position a robot's end effector at a desired location and orientation in space. The inverse of forward kinematics (which calculates end effector position from known joint angles). IK problems generally have multiple solutions or no solution.
Light Detection and Ranging. A sensing technology that measures distance by timing how long it takes a laser pulse to return after bouncing off a surface. Rotating LiDAR units generate dense 3D point clouds of the surrounding environment. Widely used in autonomous vehicles and mobile robots for navigation and mapping.
The process of determining a robot's position and orientation within a known or unknown environment. Often combined with mapping in SLAM. Approaches include probabilistic methods (particle filters, Kalman filters) and map-matching algorithms.
A flexible middleware framework for robot software development. Despite the name, ROS is not an operating system but a collection of tools, libraries, and conventions that simplify creating complex robotic systems. ROS 2 is the current version, used in research and increasingly in commercial deployments.
A machine learning paradigm where an agent learns by interacting with an environment and receiving rewards or penalties based on its actions. In robotics, RL is used to train control policies — the agent (robot) learns which actions to take in which situations by maximising accumulated reward over time.
The process of combining data from multiple sensors to produce a more accurate, complete, or reliable output than any individual sensor could provide alone. Common in robotics where cameras, LiDAR, and IMUs are fused to produce spatial understanding.
Simultaneous Localisation and Mapping. The computational problem of constructing or updating a map of an unknown environment while simultaneously keeping track of a robot's location within it. One of the foundational challenges in mobile robotics, with numerous algorithmic approaches (EKF-SLAM, particle filter SLAM, graph-based SLAM).
A rotary actuator that allows precise control of angular position, typically using a feedback mechanism (encoder) to confirm the actual position matches the commanded position. The workhorse of robotic joint actuation for lighter loads where hydraulic power is not required.