Connecting to sources…

Stanford Pupper v3 · Open-source quadruped robot dog with brushless joints (12 DoF, 3 kg, Raspberry Pi 5)

Open-source quadruped robot dog with brushless joints (12 DoF, 3 kg, Raspberry Pi 5) · Hands-On Robotics (501(c)(3) nonprofit) · Stanford CS 123 course team · …

Maker
Hands-On Robotics (501(c)(3) nonprofit) · Stanford CS 123 course team
Form
Open-source quadruped robot dog with brushless joints (12 DoF, 3 kg, Raspberry Pi 5)
For
University AI-robotics courses (Stanford CS 123), self-builders, and companion programmes in hospitals and clinics
Sensors
BNO086 IMU, IMX296 222° fisheye camera in the nose, a microphone for voice, motor drivers reporting joint angle, velocity and effort, battery-voltage ADC (tech specs page); the BOM lists the camera, microphone and speaker as optional
Connectivity and app
Driven with a PS5 wireless controller; reached over SSH, keyboard and mouse, or a Wi‑Fi hotspot hosted by Pupper, with Foxglove Studio for visualisation; AI mode calls OpenAI GPT‑Realtime, Google Gemini and Cartesia cloud APIs over the internet (docs)
Subscription
Open-source project, no subscription; AI features need your own OpenAI, Google and Cartesia API keys, and their cost is not stated
Openness
pupperv3-monorepo (ROS 2 control, neural gait controller, AI agent) and pupperv3-mjx are GPL‑3.0, pupper_v3_description is MIT; Onshape and Fusion 360 CAD, manufacturing files for both PCBs and the BOM are public, and the docs say “everything is open source” without naming a hardware licence
Privacy
Nose camera and microphone; in AI mode speech streams to OpenAI, camera frames go to Google Gemini and speech is synthesised by Cartesia's cloud (docs); the install guide prints the image's default password.
Price
USD 3400 (2026-11-22)
Not stated
Recommended age, assembly time, hardware CAD and PCB licence, runtime (the docs give only a 40-minute limit), AI cloud cost

Vendor claims

  • One hour of simulated training equals three months of experience (reinforcement learning)
  • Designed for use around people, including children: covered pinch points, limited torque, an E‑STOP on its back
  • A curriculum from years of Stanford teaching, from motor control to visual RL and LLM agents
  • Fully open source; about $2000 to build yourself

Parts

  • Raspberry Pi 5 · 8GB main computer running the robot system
  • SC1112 · Runs robot control and vision software
  • IMX296 · Captures fisheye images in the nose
  • m12x-p05 · Holds the camera lens
  • m32076m20 · Provides a fisheye field of view for the camera
  • BNO086 · Measures six-degree-of-freedom orientation and motion
  • GIM4305 · Drives leg joints for three degrees of freedom
  • PS5 · Controls robot motion wirelessly
Parts and sources

Sources