← Paper Digest

Stanford Doggo: An Open-Source, Quasi-Direct-Drive Quadruped

Nathan Kau, Aaron Schultz, Natalie Ferrante, Patrick Slade

Seminal May 2019

TL;DR Stanford Doggo is a 4.8 kg open-source quadruped driven by 3:1 belt-drive quasi-direct-drive actuators that costs under $3000 to build. It jumps 1.14 m and reaches a vertical jumping agility of 2.23 m/s, 22% above the previous best robot.

What they do

State-of-the-art legged robots take years to develop and are usually expensive custom designs. Stanford Doggo is a small quadruped (42 cm long, 4.8 kg, 8 degrees of freedom) built around a quasi-direct-drive (QDD) belt transmission. The CAD files, bill of materials, wiring and software are open-source, and the robot needs only hand tools to assemble.

How it works

  • Coaxial belt drive. Each leg is a parallel linkage driven by two motors through 3:1 belt drives with 3D-printed plastic pulleys. The two drive shafts nest coaxially, so the leg can rotate continuously, unlike Minitaur’s.
  • Why QDD. By the paper’s motor scaling laws, a transmission with ratio N saves mass over a larger direct-drive motor if it weighs less than (sqrt(N) - 1) times the motor mass. The belt drive weighs 27% of the motor mass.
  • Electronics. A Teensy 3.5 sends leg position commands at 100 Hz to four ODrive controllers, which run field-oriented control at 10 kHz on T-Motor MN5212 motors.
  • Control. Walking, trotting, bounding and pronking come from open-loop piecewise-sinusoidal foot trajectories, tracked with PD control on leg angle and extension that acts as virtual stiffness and damping.

Results

  • Against Minitaur’s direct-drive actuator: 76% higher continuous torque (1.51 vs 0.86 Nm) and 37% higher peak torque for 8% more mass and about three times the reflected inertia, at $120 instead of $280.
  • Foot force estimated from motor currents tracks a load cell with 2.8% mean absolute error on steps and 5.3% on sine waves; control bandwidth is about 150 Hz.
  • Maximum jump height is 1.14 m, more than twice that of any other quadruped compared. Vertical jumping agility is 2.23 m/s, against 1.83 m/s for Salto-1P, matching the galago.
  • Steady velocity is 0.9 m/s with a cost of transport of 3.2.

Limitations

All gaits are open-loop. The authors say instability of the open-loop tracking limited the top speed, and the cost of transport is higher than Minitaur’s (2.3). The QDD actuator has 2.7 times Minitaur’s reflected inertia, and the belts needed occasional re-tensioning. Closed-loop gait controllers are left to future work.

Figures

From Figure 1: Stanford Doggo, an open-source, quasi-direct-drive quadruped robot.
From Figure 1: Stanford Doggo, an open-source, quasi-direct-drive quadruped robot.
From Figure 2: Assembly of the mechanical components in one leg. Red and blue colors indicate parts corresponding to ...
From Figure 2: Assembly of the mechanical components in one leg. Red and blue colors indicate parts corresponding to one belt drive from motor to leg linkage.
From Figure 4: Control framework for the Stanford Doggo.
From Figure 4: Control framework for the Stanford Doggo.
From Figure 6: A comparison of the force applied by the leg as measured by a load cell and the force estimated from t...
From Figure 6: A comparison of the force applied by the leg as measured by a load cell and the force estimated from the kinematics and motor currents.
Abstract

This paper presents Stanford Doggo, a quasi-direct-drive quadruped capable of dynamic locomotion. This robot matches or exceeds common performance metrics of state-of-the-art legged robots. In terms of vertical jumping agility, a measure of average vertical speed, Stanford Doggo matches the best performing animal and surpasses the previous best robot by 22%. An overall design architecture is presented with focus on our quasi-direct-drive design methodology. The hardware and software to replicate this robot is open-source, requires only hand tools for manufacturing and assembly, and costs less than $3000.

Summaries are written automatically from each paper's text. Figures and videos come from the paper and its project page; all credit goes to the authors.