Eight weeks, one robot, one layer at a time
One 3-hour session a week, in person in Menlo Park. Each week adds a new field of engineering to the same machine, so nothing is learned in isolation.
The tags on the right show which fields of engineering are in play that week. They only ever add up.
Robot architecture, lab safety & your first limb
Understand the robot as a system before touching a screw: degrees of freedom, and the loop between actuators, sensors, controller, and battery. Start an engineering notebook. Then build one limb, which teaches joints, servo orientation, fasteners, and alignment.
Arms & legs
Split into sub-teams to build the remaining limbs. Cable routing, bearings, torque limits, and checking every joint moves freely before moving on.
Torso, head & neck
Integrate the limbs onto the torso, then install the head, neck, and stereo cameras. Finish with a mechanical inspection checklist.
Electronics & wiring checks
Power system, onboard computer, servo bus, IMU, connector labeling, and multimeter basics. Every connection is checked before power-on.
Hard lesson: bad wiring beats good code every time.Safe power-on, servo IDs & calibration
Follow the power-on checklist, assign servo IDs, and zero-point calibrate every joint.
Milestone: the robot powers on safely, and every joint is recognized and calibrated.Joint scripts & logging
Linux, the Python environment, and robot config files. The practical task: move joint 3 by 10 degrees, log the commanded vs. actual angle, and explain the difference. A real robotics lesson, not AI magic.
Keyframe motion & MuJoCo intro
Open-loop vs. closed-loop control. Design a simple standing or keyframe motion and run it on the robot. Load the robot’s model in MuJoCo and compare one motion in simulation and on hardware.
Reinforcement learning preview & showcase
What a policy, reward, observation, and action are. Watch a supplied policy train in simulation and change one reward. Then the showcase: a demo of your robot’s verified motion and a short team presentation, including what didn’t work and why.
Milestone: the showcase.Three hours, mostly hands on the robot
Every session follows the same rhythm, so students spend their energy on the robot, not on figuring out what to do.
A short primer
Just enough theory to know what you’re about to try, and what to measure.
Hands on
Build, wire, run, break, fix. Teammates rotate between operating the robot and checking each other’s work.
Debrief
What did you predict? What actually happened? Why? It all goes in the engineering notebook.
Debugging isn’t a detour. It’s the curriculum.
Our approach is described in a peer-reviewed paper accepted at IEEE TALE 2026. Read it ↗
One robot connects every subject
When a standing attempt fails, students explain it using both balance and control. Mechanical, electrical, and software work all show up in the same behavior.
Big tasks, checkable steps
One actuator, one limb, the lower body, the full robot. Each has an explicit success check: safety induction, component checks, power-on, calibration, first motion.
Predict in simulation, check on hardware
Students make predictions in MuJoCo, then measure the real robot, and explain the difference.
Worked example → one change → your call
Start from working code, change one thing, then make bounded decisions of your own, like designing your team’s keyframe motion.
Team robot, individual understanding
Four students share one robot, but each keeps their own notebook of predictions, measurements, and failure explanations.
Same motion, two worlds
On the left, the robot’s model walking in simulation. On the right, a walking trial on real hardware.
Simulation · MuJoCo
Hardware · walking trial
Going further: the 22-week research track
The same platform, much deeper: model matching in MuJoCo, control, reinforcement-learning experiments, and a technical write-up. Ask us about availability.
8 WEEKS
Summer intensive
Reproduce ToddlerBot in person, about 6 hours a week, mostly on weekends: assembly, wiring, and bring-up.
12 WEEKS
Fall academy
Simulate, control, and improve. Weeknight virtual seminars and code reviews, plus in-person lab days for sim-to-real testing.
2 WEEKS
Output sprint
A portfolio-quality final project: demo video, research-style write-up, GitHub docs, a short talk, and a reflection on what failed.
Everyone learns the whole stack. The final project goes deep in one direction: