Metal2Humanoid

MIT IAP course · 4–29 Jan 2027 · MIT, Cambridge, MA · Taught by Ben Zandonati, Willie McClinton, Seth Robles, James Kim

Building a Humanoid Robot from the Ground Up

A hands-on introduction to humanoid robotics, tracing the path from motors and actuators to coordinated motion and learned behavior. Over four weeks, students work in teams to assemble and test components of an open-source humanoid based on Berkeley Humanoid Lite. Topics include motors, transmissions, encoders, robot kinematics, feedback control, simulation, reinforcement learning, and transferring policies to physical hardware.

Each week combines two lectures, a guided hardware lab, and a scaffolded Python assignment. Projects progress from actuator position control to limb motion, simulated locomotion, and system integration, culminating in a team demonstration and brief technical report. Hardware components and software scaffolds are prepared in advance.

What we’re building

The robot the course builds is a version of the Berkeley Humanoid Lite, an open-source humanoid made from 3D-printed parts and off-the-shelf motors. Here it is in simulation, in your browser, walking with an RL-trained policy — the same kind of controller students train in week 3.

▶ Run the simulationRuns in your browser · about 4 MB

Teaching team

• Ben Zandonati: instructor, MIT EECS
• Willie McClinton: co-instructor, curriculum and robotics instruction
• Seth Robles: co-instructor and undergraduate laboratory teaching assistant
• James Kim: co-instructor and hardware laboratory instructor

Faculty sponsor: Tomás Lozano-Pérez, MIT EECS.

Who it’s for

Open to undergraduate and graduate students. An introduction to computer science or robotics, or equivalent experience, is expected. Assignments use Python. No prior hardware assembly, control theory, or machine learning experience is required.

Learning objectives

• Explain how motors, transmissions, encoders, and feedback control produce controlled joint motion.
• Model robot limbs and implement basic end-effector motion in simulation.
• Train and evaluate a scaffolded locomotion policy and identify simulation-to-reality challenges.
• Integrate and test robot subsystems, document results, and demonstrate coordinated behavior.

Schedule

Lectures are on Tuesdays and Thursdays, 1:00–2:30 p.m., from 5 to 28 January; hardware labs on Fridays, 1:00–2:30 p.m., from 8 to 29 January. There’s an optional office hour each week. Rooms are to be confirmed.

Open a week for its lectures, lab and assignment.

Participation and the final demonstration

Activities include four weekly assignments, laboratory participation, and a final team demonstration and report. Instructors give feedback on understanding, testing, and documentation. The course is non-credit, with no formal grade. Untethered walking and manipulation are stretch goals, and the simulation-based work keeps everyone learning when the hardware needs troubleshooting.

Lectures

1. Motor fundamentalsComing soon · 5 Jan 2027
2. Transmissions, encoders, and feedback controlComing soon · 7 Jan 2027
3. Robot modeling and forward kinematicsComing soon · 12 Jan 2027
4. Inverse kinematics and trajectory generationComing soon · 14 Jan 2027
5. Reinforcement learning for locomotionComing soon · 19 Jan 2027
6. Simulation-to-reality transferComing soon · 21 Jan 2027
7. Skill representations and sequencingComing soon · 26 Jan 2027
8. Integrated robot behaviorComing soon · 28 Jan 2027

@online{zandonati2027metalhumanoid,
  author    = {Ben Zandonati and Willie McClinton and Seth Robles and James Kim},
  title     = {Metal2Humanoid},
  year      = {2027},
  month     = {jan},
  url       = {https://benzandonati.co.uk/notes/courses/metal2humanoid/},
  urldate   = {2026-11-05},
  note      = {Online; accessed 2026-11-05},
  keywords  = {humanoids, robotics, course},
}