Miniature Sandbox
for Embodied AI

ShuttleCore-Mini is a 35cm tall, 23-DoF desktop-level humanoid robot platform engineered for rapid sub-scale prototyping and agile hardware iteration. Built to systematically eliminate Sim2Real physics and timing blind spots before scaling to full-size humanoids.

Modular Product Line

Modular Architecture

Every component is designed as an independent unit that snaps into a larger system — from a single leg to a complete humanoid robot.

Minimal Unit

Standalone mechanical modules — legs, arms, and heads with full degrees of freedom, ready for independent deployment or system integration.

View Modules →

Subsystem

Paired assemblies with integrated control algorithms — bipedal locomotion platforms and dual-arm upper body systems.

View Subsystems →

Full Robot

The complete modular humanoid — all subsystems combined into a fully functional robot with 26+ degrees of freedom.

View Full Robot →

Kinematics

Spatial Topology

Architecting a rigid sandbox to eliminate spatial play and non-linear distortion, ensuring absolute geometric integrity under full-dynamic running conditions.

Absolute Root Anchor

Locks the primary reference frame at the geometric center of the pelvis. This strategic placement permanently isolates lower-body state tracking from upper-body attitude feedback contamination.

Coordinate Lock

Utilizes a 1-DoF pitch-only neck to bypass complex multi-axis spatial transformations. This structural constraint offloads redundant computing overhead while maintaining deterministic visual tracking.

Orthogonal Virtual Mapping

Executes a real-time kinematics mapping matrix for the non-linear 2-DoF parallel ankle. Translates complex pushrod constraints into transparent linear outputs for the RL tracking policy.

Follower Decoupling

Strips the peripheral elbow linkages from the core torque-control pipeline. By degenerating these axes into pure position followers, the network eliminates distal inertia to protect upper-body stability.

Control

Heterogeneous Computing

Isolating compute pipelines to secure maximum execution bandwidth, blocking systemic processing latency and voltage drops simultaneously.

Hardware VPU Offloading

Fully offloads depth processing and point-cloud generation to the dedicated hardware VPU. This isolation secures critical tensor bandwidth on the primary compute core.

Localized Policy Inference

Deploys the real-time reinforcement learning policy and terrain adaptation locally. Eliminates off-board transmission latency to guarantee deterministic control loops.

Dual-Track Backbone

Isolates peripheral communication via a dedicated high-bandwidth dual-track CAN-FD network. Prevents joint telemetry congestion and cross-talk during full-dynamic maneuvers.

Transient Power Shield

Integrates a high-capacity solid-state capacitor array to defend against voltage brownouts during peak current draw, ensuring zero-reset reliability.

Ecosystem

Ecosystem Strategy

Accelerating open-source community adoption while safeguarding high-margin hardware manufacturing monopolies.

Digital Twin Sim

THE OPEN LAYER

Structural DLCs