Skip to content
buildrobots.io
LearnStart

Path 3 of 3 · Hu · 4 steps

Humanoid

Everything below at once, standing up. Small biped first, then a full humanoid with a published package.

  1. 01

    A desk-scale biped

    Bipedal bring-up at a size where falls are cheap.

    Done whenA small biped stands, then walks with a learned policy, and you have a fall routine.

    Read / watch / run

    • Open Duck Mini

      Antoine Pirrone · Repo · free · github.com/apirrone/Open_Duck_Mini

      A small, printable bipedal robot with an RL walking policy: legged bring-up at desk scale.

    • MuJoCo documentation

      Google DeepMind · Docs · free · mujoco.readthedocs.io/en/stable/overview.html

      The physics engine behind most open robot sims: MJCF models, actuators, contacts, sensors.

    BenchOpen Duck Mini parts: printed shell, small servos, an IMU and a single-board computer, per its repo BOM.
  2. 02

    Build a humanoid

    Dozens of actuators on one bus, harnessing, safe power-up.

    Done whenEvery joint homes and holds; the e-stop cuts power; the robot holds a supported stand.

    Read / watch / run

    • Berkeley Humanoid Lite docs

      UC Berkeley Hybrid Robotics · Docs · free · berkeley-humanoid-lite.gitbook.io/docs

      Build an open, printed-gearbox humanoid: actuators, electronics, assembly and RL locomotion.

    • ODrive getting started

      ODrive Robotics · Docs · free · docs.odriverobotics.com/v/latest/guides/getting-started.html

      Wiring, configuring and calibrating a BLDC motor driver: the core of every quasi-direct-drive leg.

    Practise on · computed

    Buildable robots of this class, easiest first by the difficulty index; robots with a build package listed first.

  3. 03

    Locomotion policies

    Run and check published walking policies, then train your own.

    Done whenA published policy walks your model in sim with the same observation contract as its source.

    Read / watch / run

    • unitree_rl_gym

      Unitree Robotics · Repo · free · github.com/unitreerobotics/unitree_rl_gym

      Train, sim2sim-check in MuJoCo and deploy walking policies for Unitree G1, H1 and Go2.

    • Isaac Lab documentation

      NVIDIA · Docs · free · isaac-sim.github.io/IsaacLab/main/index.html

      GPU robot-learning framework: environments, RL and imitation workflows, sim-to-real.

    • MuJoCo Playground

      Google DeepMind · Tool · free · playground.mujoco.org

      GPU-accelerated RL environments for quadrupeds, humanoids and hands, with sim-to-real examples.

    • Underactuated Robotics

      Russ Tedrake, MIT · Book · free · underactuated.mit.edu

      Walking, balance and control of robots that can fall over: the theory behind legged locomotion.

    Policy zoo on humanoidrobots.training
    BenchThe robot you built in the build step of this path.
  4. 04

    Hands and manipulation

    Grasping, teleoperation and imitation learning on a whole body.

    Done whenA recorded dataset of one bimanual task and a policy that repeats it.

    Read / watch / run

    • Robotic Manipulation

      Russ Tedrake, MIT · Book · free · manipulation.mit.edu

      Perception, planning and control for picking things up, with runnable notebooks.

    • LeRobot documentation

      Hugging Face · Docs · free · huggingface.co/docs/lerobot/index

      Install LeRobot, record datasets, train and evaluate policies on low-cost arms.

    • Robotics course

      Hugging Face Learn · Course · free · huggingface.co/learn/robotics-course

      Robot learning from the ground up with LeRobot: data, policies, real hardware.

    BenchThe robot you built in the build step of this path.

Resource links checked 2026-09-25.