The ideas behind robot perception, localization, and recognition, connected in a progression from intuition to mathematics and implementation.
This is an editorially selected reading order. Begin at step 1 or go directly to the topic you need.
Connect sensing, localization, mapping, planning and control, then choose a reading path for the system you want to build.
ROS 2 is a robotics middleware that splits sensing, perception, planning, and control into separate processes and connects them via DDS communication. This article covers nodes, topics, services, actions, QoS, TF2, lifecycle, implementation and debugging, along with the differences from ROS 1 and real-world operational cautions, with diagrams.
Dead reckoning is the foundational technique of updating a robot's pose and position by integrating wheel rotation, IMU readings, and velocity sensors. This article covers differential-drive kinematics, accumulating error, EKF-based correction, and implementations in indoor transport robots and autonomous driving, with equations and diagrams.
A camera infers distance indirectly from patterns of reflected light; a LiDAR measures distance directly from the time of flight of a laser pulse. LiDAR-SLAM is the technology built on registering the point clouds this active ranging sensor returns. This article works systematically from the two pillars of scan matching, ICP and NDT, through LOAM's feature design, Loop Closure and Graph SLAM, up to the lineage of landmark algorithms — LeGO-LOAM, Cartographer, LIO-SAM, FAST-LIO2.
A camera is strong on meaning, a LiDAR is strong on geometry, an IMU is strong on motion — and each has its own blind spot. From the probabilistic and optimization-based math that fuses them (Kalman Filters, Factor Graphs) to concrete pairings like VIO (Camera×IMU), LIO (LiDAR×IMU), and BEV Fusion (Camera×LiDAR), a systematic tour of how Sensor Fusion turns several sensors into a single world model.
Diagnose fusion failures through timestamps, coordinate frames, mounting geometry and covariance.
A camera alone, or a LiDAR alone, breaks down in certain situations. Combining either sensor with an IMU (accelerometer + gyroscope) covers for that weakness -- this is what VIO (Visual-Inertial Odometry) and LIO (LiDAR-Inertial Odometry) do. Starting from why an IMU alone cannot recover position, this article works through the two design philosophies -- Filter-based and Optimization-based -- and the landmark algorithms from ROVIO, OKVIS, VINS-Mono, and OpenVINS through LIO-SAM and FAST-LIO2.
Use synthetic trajectories to distinguish absolute and relative errors, then define matching, alignment and failure reporting for an evaluation.
Forward kinematics finds the hand's position and orientation from the joint angles; inverse kinematics solves the reverse problem; and the Jacobian bridges the two through "velocity." From building coordinate transforms with DH parameters, through analytic IK, numerical IK (Damped Least Squares), singularities, and redundant degrees of freedom — a ground-up, systematic account of the kinematics that governs how a robot arm moves.
Starting from what actually happens the instant a position-controlled robot arm touches a rigid object — and why that makes assembly, polishing, and human-contact tasks impossible — this article works through impedance control, which links force and displacement via a virtual spring, damper, and mass; admittance control, which reverses that causality; and hybrid position/force control, which separates position and force by spatial direction, building the ideas up from equations and diagrams.
Why can a small input move a hydraulic excavator's massive arm? This article treats the pump, control valves, cylinder, hydraulic fluid, sensors, and controller as a single closed loop, covering the pressure/flow/efficiency equations, current technology in construction and agricultural machinery, and electro-hydraulic conversion and current research, with diagrams.
ArduPilot is an open-source autopilot that implements a vehicle's sensor fusion, attitude/position control, navigation missions, and failsafes. This article covers the differences between Copter, Plane, and Rover, plus EKF, PID, MAVLink, SITL, and real-vehicle safety testing, with diagrams.
Swarm control solves problems a single robot can't, using nothing but "simple local rules" applied across many robots. From reproducing flocking with Boids, through consensus algorithms for agreement, to the three approaches to formation control — leader-follower, virtual structures, and behavior-based — this article systematically organizes the field around the divide between centralized and decentralized architectures.