Robotics Institute

Fundamentals

What is a Robot?

A robot is a programmable machine capable of sensing its environment, making decisions, and acting on the physical world. Modern robotics integrates mechanical engineering, electrical engineering, and computer science — combining kinematics, dynamics, control theory, perception, and planning into autonomous systems.

Key Concepts

Kinematics — the study of motion without forces: forward kinematics (joint angles to end-effector pose) and inverse kinematics (desired pose to joint angles)
Dynamics — how forces and torques produce motion; Newton-Euler and Lagrangian formulations
Control Theory — PID, optimal control, and model predictive control to make robots follow desired trajectories
SLAM — Simultaneous Localization and Mapping; building a map of the environment while tracking the robot's position within it
Motion Planning — finding collision-free paths through configuration space (RRT, PRM, A*)
Grasping & Manipulation — how robots pick up and interact with objects; force closure, grasp planning, dexterous manipulation

Robotics vs Automation

Automation follows fixed, pre-programmed sequences. Robotics involves sensing, reasoning, and adapting to uncertainty.
A robotic system must perceive its environment (cameras, LiDAR, IMUs), plan actions, and execute them with feedback control.
Modern robots combine classical control with machine learning for tasks like grasping novel objects or navigating unstructured terrain.
Key challenge: the real world is messy — robots must handle noise, uncertainty, and unexpected situations.

Free Courses

MIT OCW: Introduction to Robotics

Covers mechanics, dynamics, and control of robot manipulators. Includes lecture notes, problem sets, and labs.

MIT OpenCourseWare | Free | Intermediate

Stanford CS223A: Introduction to Robotics

Oussama Khatib's classic course. Kinematics, dynamics, control, motion planning. Video lectures available.

Stanford Engineering Everywhere | Free | Intermediate

Coursera: Robotics Specialization (UPenn)

Six-course series covering aerial robotics, perception, estimation, planning, and capstone. Free to audit.

University of Pennsylvania | Free audit | Beginner to Advanced

Coursera: Modern Robotics (Northwestern)

Based on the Lynch & Park textbook. Rigorous treatment of kinematics, dynamics, planning, and manipulation.

Northwestern University | Free audit | Intermediate

ROS 2 Tutorials

Official tutorials for Robot Operating System 2. Learn the de facto standard middleware for robotics development.

Open Robotics | Free | Beginner to Intermediate

Udacity: AI for Robotics (Sebastian Thrun)

Localization, tracking, PID control, SLAM, and path planning. Taught by the leader of the Stanford autonomous car team.

Udacity | Free | Intermediate

Textbooks

Siciliano et al: Robotics — Modelling, Planning and Control — Comprehensive graduate-level textbook. Covers kinematics, dynamics, motion planning, and force control.
Lynch & Park: Modern Robotics — Free PDF available. Rigorous, modern treatment of mechanics, planning, and control. Companion Coursera course.
Corke: Robotics, Vision and Control — Practical approach with MATLAB toolbox. Covers manipulators, mobile robots, and vision.
Thrun, Burgard & Fox: Probabilistic Robotics — THE textbook for probabilistic state estimation, SLAM, localization, and decision-making under uncertainty.
LaValle: Planning Algorithms — Free online. Definitive reference for motion planning, sampling-based algorithms, and decision-theoretic planning.

Key Research Papers

Brooks (1986): A Robust Layered Control System

Introduced the subsumption architecture — reactive, behavior-based robotics without central world models. Revolutionized how robots are programmed.

Rodney Brooks | MIT AI Lab

Khatib (1986): Real-Time Obstacle Avoidance

Artificial potential field method for real-time motion planning. Attractive fields to goal, repulsive fields from obstacles. Foundational for reactive navigation.

Oussama Khatib | Stanford

Raibert (1986): Legged Robots That Balance

Pioneered dynamic legged locomotion. Showed that running and hopping could be decomposed into three decoupled controllers. Foundation for Boston Dynamics.

Marc Raibert | MIT Leg Lab

Kavraki et al (1996): Probabilistic Roadmaps

Probabilistic Roadmap Method (PRM) for multi-DOF motion planning. Sample random configurations, connect collision-free neighbors. Still widely used.

Kavraki, Svestka, Latombe, Overmars

LaValle (1998): Rapidly-Exploring Random Trees

RRT algorithm for single-query motion planning. Incrementally builds a space-filling tree. Foundation for RRT* and modern planning.

Steven LaValle | University of Illinois

Levine et al (2018): Learning Hand-Eye Coordination

Large-scale deep learning for robotic grasping. 800,000 grasp attempts across multiple robots. Showed end-to-end learning from pixels to motor torques.

Sergey Levine | Google Brain / UC Berkeley

Open-Source Tools

ROS 2 (Robot Operating System)

The de facto standard middleware for robotics. Pub/sub messaging, hardware abstraction, package management, visualization.

C++ / Python | Open Robotics | Apache 2.0

Gazebo

3D robotics simulator with physics, sensors, and ROS integration. Test robots in realistic environments before deploying on hardware.

C++ | Open Robotics | Apache 2.0

MuJoCo

Multi-Joint dynamics with Contact. High-performance physics engine for robotics research. Now free and open-source (Google DeepMind).

C / Python | Google DeepMind | Apache 2.0

PyBullet

Physics simulation for robotics, games, and ML. Python bindings, OpenAI Gym integration, URDF/SDF support.

C++ / Python | Erwin Coumans | zlib

NVIDIA Isaac Sim

GPU-accelerated robotics simulation built on Omniverse. Synthetic data generation, domain randomization, ROS/ROS2 bridge.

NVIDIA | Free for individual use

Drake

Model-based design and verification for robotics. Multibody dynamics, optimization-based planning, and control. Used by Toyota Research.

C++ / Python | MIT | BSD 3-Clause

Industry & Hardware

Boston Dynamics

Spot (quadruped), Atlas (humanoid), Stretch (logistics). The most advanced dynamic robots in the world.

Legged robots | Hyundai subsidiary

Universal Robots

Collaborative robot arms (cobots). UR3e, UR5e, UR10e, UR16e. Industry standard for lightweight industrial automation.

Cobots | Teradyne subsidiary

Arduino

Open-source microcontroller platform. The easiest way to start building robot hardware. Huge ecosystem of sensors, motors, and shields.

Microcontrollers | Open-source hardware

Raspberry Pi

Single-board computer for robot brains. Run ROS, computer vision, and ML inference. GPIO for sensor/actuator interfacing.

SBC | Raspberry Pi Foundation

NVIDIA Jetson

GPU-powered edge computing for robotics. Run neural networks on robots. Jetson Nano, Orin Nano, AGX Orin.

Edge AI | NVIDIA

DJI RoboMaster

Educational robot with omnidirectional wheels, vision, and programmable AI. Great for learning competition robotics.

Educational robot | DJI

Reference

Kinematics & Dynamics

Forward Kinematics — compute end-effector position/orientation from joint angles using DH parameters or product of exponentials
Inverse Kinematics — find joint angles that achieve a desired end-effector pose; analytical (closed-form) or numerical (Jacobian-based)
Jacobian — maps joint velocities to end-effector velocities; used for differential kinematics, singularity analysis, and force control
Euler-Lagrange dynamics — derive equations of motion from kinetic and potential energy: M(q)q'' + C(q,q')q' + g(q) = tau

Control

PID Control — proportional-integral-derivative; the workhorse of industrial robotics. Tune Kp, Ki, Kd for stability and performance.
Computed Torque Control — model-based: cancel nonlinear dynamics, then apply linear control. Requires accurate dynamic model.
Model Predictive Control (MPC) — optimize control inputs over a receding horizon. Handles constraints. Used in legged locomotion.
Impedance Control — control the dynamic relationship between force and motion. Essential for safe human-robot interaction.

Perception & Planning

SLAM — simultaneous localization and mapping. EKF-SLAM, particle filter SLAM, graph-based SLAM (g2o, GTSAM).
Motion Planning — RRT, RRT*, PRM, A*, Dijkstra in configuration space. Sampling-based vs grid-based vs optimization-based.
Computer Vision — object detection, pose estimation, depth sensing, visual servoing. OpenCV, deep learning (YOLO, SAM).
Grasping — force closure analysis, grasp planning, contact modeling. GraspIt!, Dex-Net for data-driven grasping.