Stone Industries / CASA

CASA is in active in-public analysis. The case study below summarizes the current thesis; the full body of work lives in the public analysis repo.

Core thesis

The hardware problem of humanoid robots was already solved when ChatGPT came out. ChatGPT was the missing piece of the robot: the brain.

The four-pillar bet:

1. Classical-first — IK, MPC, pathfinding, parametric grasp catalogs do the precise mechanical work
2. Tactile-first — fingertip ToF sensors close the contact loop where vision can’t
3. Scene graph — persistent semantic understanding of the environment
4. LLM-as-brain — high-level reasoning, task decomposition, conversational interaction

CASA — Classical Autonomous Servant Architecture — targets the consumer indoor fetch use case rather than the manufacturing use case most humanoid companies are chasing. The market positioning, sensor architecture, and locomotion approach all derive from that scope choice.

Roadmap

• 2026 — ship CASA Sim, a Unity-based bootstrap product running the real control stack as a Steam/console game
• 2028 — first hardware prototype
• 2029 — consumer launch

The bootstrap-via-sim path lets the company ship a product, generate revenue, and accumulate a real-environment dataset before any hardware exists.

The full analysis

The complete 9-document analysis covers:

• Technical thesis (canonical) — the four pillars reconciled
• Master plan — 8-layer architecture, hardware roadmap, team, monetization
• Sensor and perception architecture — fingertip ToF, head LiDAR/radar/UWB, the industry sensor-camp split
• Classical-first thesis + founder analogues — Walton, Musk, Huang, Kelleher case studies
• Competitive intel — Figure / 1X / Tesla / Boston Dynamics reads
• Market positioning — why humanoids lose at manufacturing
• Locomotion and modular feet — articulated-foot moat, sim fidelity as the real blocker
• Simulation tooling — Unity vs Isaac Sim decision

Build-in-public is the strategy: the analysis lives openly so the thinking is testable.