SOLSTICE · HORIZON BA9B

Build a useful bridge between a real six-legged robot, aerospace engineering study and software development while keeping the robot reachable only over a private network. The system also needed to expose sensor and controller state clearly, so that GNC experiments could be inspected rather than hidden behind an “AI autopilot” claim.

Use a Raspberry Pi API as the robot-side boundary; connect the Freenove motion server locally; read MPU6050 telemetry for Astro’s own attitude estimator; process attitude through a multiplicative/error-state EKF, rate-limited roll/pitch guidance and discrete LQR; send bounded body-attitude requests through Freenove kinematics. A private Django bridge exposes only owner-authorized telemetry and fixed, bounded actions. Tutor adds reviewed telemetry to learning questions; Codex inspects authorized GitHub text and requires a separate confirmation for a draft PR.

Owner-run tests have demonstrated servo movement, manual forward gait and yaw rotation, private Pi connectivity, IMU telemetry, gyro calibration, GNC mode changes and attitude-command output. Reported GNC telemetry reached approximately 25 Hz. Physical stabilization is still experimental: tremor on a level surface and occasional LQR saturation remain unresolved. The 3D view uses a reference hexapod model, not an exact FNK0052 digital twin; the project makes no claim of autonomous slope walking or flight readiness.