Gray
quadruped

Gray is a learning platform for the coupled problems inside legged robotics. A quadruped cannot be separated neatly into mechanical and software halves: link geometry affects reachable motion, actuator choices affect control bandwidth, and the simulated robot must reflect the physical build closely enough for experiments to mean something.

The project uses a 3D-printed body to make iteration accessible. Components can be revised, manufactured, assembled, and tested without waiting for specialized production, which supports the project's experimental character.

Development spans CAD and packaging, printed-part iteration, actuator integration, kinematic reasoning, simulation, and control experiments. Mechanical interfaces are designed with wiring, joint range, assembly sequence, and replacement in mind rather than only exterior form.

On the software side, the repository provides a shared place for simulation and reinforcement-learning work. The aim is to explore locomotion while keeping an eye on the simulation-to-hardware gap: joint limits, mass distribution, friction, latency, and actuator behavior all influence whether a learned policy can transfer.

Gray is explicitly a work in progress. The current outcome is a common platform on which mechanical and controls decisions can be tested together, not a claim of finished autonomous locomotion.

The public repository records the evolving work and makes the project inspectable. As the platform changes, versioned geometry, parameters, and test observations will be important for relating software behavior to the specific physical configuration.

Robotics exposes interface mistakes quickly. A small packaging compromise can restrict a joint; an uncertain mass property can distort simulation; a controller can appear successful while exploiting behavior the hardware cannot reproduce.

The next work is centered on disciplined iteration: improve the physical model, validate basic joint behavior, establish repeatable tests, and increase locomotion complexity only after the simpler layers are understood.

• Scope and claims are limited to what the surviving project record supports.
• Future updates will add verified media, measurements, and milestones as they become available.

The public repository preserves the inspectable portion of this project.