A $200 Open-Source Quadruped Robot Dog: From Hexapod to DIY Full Build Guide

From Six Legs to Four: A Successful Hardware Slim-Down

An open-source hardware enthusiast recently shared his latest creation with the community — the Quattro ZBD quadruped robot dog. Rather than building from scratch, it evolved from his existing hexapod by "chopping off two legs." The author admits the transition went surprisingly smoothly: all the features developed for the hexapod — including body leveling, emotes, and puppet mode — migrated cleanly to the new platform.

Dropping two legs comes with real, tangible benefits. Compared to the hexapod, the Quattro ZBD is lighter, faster, and more agile in many scenarios. More importantly, the cost dropped to around $200 thanks to significantly fewer components — making it a much more accessible entry point for DIY enthusiasts interested in legged robotics.

Hardware: Built Entirely from Off-the-Shelf Modules

This robot dog uses no custom chips whatsoever. The core controller is the well-regarded ESP32 S3, paired with off-the-shelf components from the Arduino ecosystem — a "anyone can replicate this" kind of build.

For actuators, it uses 12 DS3218 high-torque servos (3 degrees of freedom per leg), coordinated by a PCA9685 servo driver board. Attitude sensing is handled by an MPU6050 IMU, which is the foundation for the body leveling feature. The power system is thoughtfully designed: a 3S LiPo battery (3000–6000mAh) feeds two 15A adjustable buck converters — one per pair of legs — while a separate 5V 3A buck converter handles logic circuitry, preventing high-current servo movements from destabilizing the microcontroller.

The complete bill of materials also includes: a voltage sensor board, a 2-inch TFT display (for status readouts or expressions), 12 × 8x5x16mm ball bearings, Wago quick-connect terminals, and approximately 100 M3x8 screws and fasteners. The parts list is fully transparent and practically ready for direct purchase.

The DS3218 is a metal-gear servo commonly used in DIY robotics, rated at approximately 21kg·cm of torque (at 6V) — more than adequate for joint actuation in small-to-medium legged robots. The PCA9685 is a 16-channel PWM driver chip from NXP that communicates with the host microcontroller via I²C, allowing a single ESP32 to precisely control up to 16 servos simultaneously without consuming GPIO pins with software PWM. The MPU6050 is a 6-axis IMU combining a 3-axis accelerometer and 3-axis gyroscope, providing real-time attitude data essential for body leveling. A 3S LiPo has a nominal voltage of 11.1V (up to ~12.6V fully charged); adjustable buck converters step this down to the servo operating voltage (typically 6–7.4V) and the 5V needed for logic, with the split-rail design effectively isolating servo current spikes from the microcontroller.

Software & Motion Control: IK Today, Reinforcement Learning Tomorrow

The current version's gait is computed by an Inverse Kinematics (IK) solver that operates independently on each leg. This is also why the author could migrate the hexapod code to the quadruped so quickly — the core mathematical framework is universal, requiring only gait mechanism adjustments. The author describes the process as "surprisingly straightforward."

That said, he isn't satisfied with the status quo. While IK-driven gaits are stable, they look somewhat mechanical. As a result, the author has revealed the next step: introducing reinforcement learning to train a control policy in place of the fixed IK solver, with the goal of producing more natural, fluid locomotion. This approach aligns with the learning-based control strategies used in mainstream commercial legged robots (such as those from Boston Dynamics and Unitree) — only implemented here on a sub-$200 hobbyist platform, which makes it particularly compelling.

The companion software is equally complete: an ESP32 control script and an Android remote control app (with APK) are both open-sourced on GitHub.

Inverse Kinematics (IK) is the classic approach to legged robot motion control: given a target position for a leg's end-effector (foot), the algorithm works backward to calculate the required joint angles. Its counterpart, forward kinematics, computes end-effector position from known joint angles. IK's advantages lie in mathematically well-defined relationships, predictable computation, and ease of debugging — making it an ideal entry-level control strategy. Its limitations, however, are clear: gait trajectories must be hand-designed, the robot has limited adaptability to uneven terrain, and motion naturalness is constrained by preset curve parameters. Reinforcement learning, by contrast, lets the robot autonomously learn control policies through extensive trial-and-error in simulation, often producing smoother gaits with greater robustness to complex terrain. The transition from IK to RL represents the classic shift from rule-based to data-driven control in legged robotics — a path both Boston Dynamics' Spot and Unitree's Go series have followed.

Fully Open-Source: All Files Available for Free

The greatest value of this project lies in its thoroughly open-source nature. The author provides a complete set of replication resources:

• CAD designs, 3D print files, and wiring diagrams (Patreon)
• MakerWorld model page (with additional build photos)
• ESP32 control scripts (GitHub)
• Android remote app source code and APK (GitHub)
• Detailed demo video (YouTube)

In short, any maker with basic 3D printing and soldering skills can follow this documentation to build their own quadruped robot dog from scratch — no expensive commercial kits required.

Next Stop: A Bipedal Version

The "2 more to go" joke in the original post title hints at the author's broader ambitions — after dropping to four legs, he's now working on a bipedal version. The updated leg structures needed for bipedal locomotion are nearly complete, allowing the robot to stand on two legs, while the rest of the components will reuse the existing design as much as possible. The author says that if he can get it walking successfully, he'll release the bipedal version within a few weeks.

From hexapod to quadruped to biped, this creator is essentially climbing the legged robotics difficulty curve step by step using the same low-cost hardware and software framework. For enthusiasts who want hands-on experience without immediately tackling the challenges of bipedal balance, this iterative approach is itself a highly instructive roadmap.