Track 09
Mobile & Legged Robotics
From differential drive to humanoid locomotion. Getting a body through the world on wheels, legs, or wings.
10 published · 0 planned · 10 lessons total
- 01→
Differential drive: the robot you build first
PublishedTwo wheels, two motor commands, endless variations. The kinematics, the command conversion, the slip problem, and the 50 lines of code that control any diff-drive robot.
~13 min
- 02→
Ackermann steering for car-like robots
PublishedCars don't pivot. They turn around an instantaneous center, with both front wheels at different angles. The kinematic constraints, the bicycle-model approximation, and what changes for planners and controllers.
~13 min
- 03→
Omnidirectional drive and mecanum wheels
PublishedThree-wheel and four-wheel omni kinematics. When you need holonomic motion, when you don't, and the mechanical price you pay for sliding sideways.
~11 min
- 04→
Quadruped locomotion: gaits and phase diagrams
PublishedTrot, bound, pace, gallop — four families of quadruped gait, each a different phase pattern. The phase diagram you'll see in every legged robotics paper, demystified.
~14 min
- 05→
ZMP, capture point, and the walking cookbook
PublishedZero Moment Point and Capture Point — the classical tools that kept Honda's ASIMO upright. Still useful in the era of learned gaits, and the lingua franca every humanoid team speaks.
~13 min
- 06→
Whole-body control for humanoids
PublishedHierarchical task control, friction cone constraints, the QP-based formulation every humanoid team uses. Turning a high-level plan into 25+ DOF of coordinated joint motion.
~14 min
- 07→
Reinforcement-learned gaits: the 2024–26 revolution
PublishedFive years ago, every quadruped used hand-tuned MPC. Today they all use RL. Here's the recipe — domain randomization, action space choices, real-world deployment — that ate the field.
~13 min
- 08→
Drone control: from PID to nonlinear MPC
PublishedThe hierarchy that flies every quadrotor — three nested loops, each tuned for its timescale. PID for the rate loop, MPC for the trajectory, and the differential-flatness trick that makes it tractable.
~12 min
- 09→
Autonomous driving stack anatomy
PublishedThe perception → prediction → planning → control pipeline, mapped one level at a time. The architecture behind every L4 vehicle, with honest notes on what's still hard in 2026.
~14 min
- 10→
Social navigation: robots around people
PublishedThe frontier where metric planning meets human behavior modeling. Personal space, intent inference, and the hard task of being a 'polite' robot in a crowded space.
~12 min