Sovereign Physical AI • Robotics • Multi-Agent Systems

FADEDMATRIX

An autonomous computing cluster and robotics laboratory exploring 3D cable kinematics, mobile manipulation, custom multi-agent governance, and continuous closed-loop learning — built purely on open-source foundations.

Core Philosophy: 100% Ad-Free, Tracker-Free, and Privacy-Respecting. This laboratory uses local GPUs, edge accelerators, and on-demand simulation arbiters to train models without third-party cloud data extraction.

Robotics & Physical AI Systems

Stringman CDPR

Ceiling-mounted 4-cable parallel robot engineered for expansive 3D workspace manipulation, anti-swing dynamics, and vision-guided object handling.

Kinematics: 4-Cable Inverse Solver • Anti-Pendulum Damping • AprilTag Fiducials

Yahboom ROSMASTER M3

Jetson Orin NX mobile base with 6-DOF robotic arm. Features custom laser scan timestamp skew compensators and camera-to-arm DH transform coordinate models.

Hardware: Jetson Orin NX • ROS 2 Humble • Mutual-Exclusion Nav2 Lease Gate

On-Demand Sim Arbiter

Elastic GPU compute scheduler dynamically scoring VRAM headroom and thermals across 88 GiB of cluster VRAM to place physics and perception simulations on-demand.

Fleet: RTX 4090s, RTX 5080, RTX 3090 • 50-Trial Grasp Loop: 100% PASS

Multi-Agent Orchestration & Operator Cockpit

BB Multi-Model Dispatcher

Custom role-to-model governance engine mapping agent personas (chief, implementer, reviewer, explorer) to optimal local and remote LLM providers under strict permission bounds.

Engine: spawn-governed.sh • bb-watch-agents.service • Multi-Machine Routing

Unified VS Code Studio

Multi-agent workspace environment with shadow cockpit contract enforcement, live diff inspection, and automated Model Context Protocol (MCP) tool integration.

Workflow: Shadow Cockpit Contracts • AST Diagnostics • MCP Governance Plane

Autonomous Error Practice & Memory

Iterative self-correction engine replaying failed tool calls and AST schema rejections until 0 errors are achieved, synthesizing real receipts into Qdrant vector memories.

Learning: DPO Alignment Pairs • Qdrant Vector DB (homelab_experience_v1)

Open-Source Attribution & Citations

We honor and promote the creators, maintainers, and researchers whose open code makes this laboratory possible:

Project / Technology Creator / Maintainer Upstream Repository License
LeRobot (Physical AI) Hugging Face (Remi Cadene, Simon Alibert et al.) github.com/huggingface/lerobot Apache 2.0
Cranebot3 & CDPR Research Nathaniel Nifong (Neufangled Robotics) github.com/nhnifong/cranebot3-firmware MIT
Hermes Agent & Models Nous Research (Teknium & Team) github.com/NousResearch Open Model / MIT
llama.cpp Engine Georgi Gerganov & Community github.com/ggerganov/llama.cpp MIT
Qdrant Vector Engine Qdrant Solutions GmbH github.com/qdrant/qdrant Apache 2.0
LiteLLM Gateway BerriAI (Krrish Dholakia, Ishaan Jaffer) github.com/BerriAI/litellm MIT
Frigate NVR Blake Blackshear github.com/blakeblackshear/frigate MIT
Home Assistant Paulus Schoutsen & Community github.com/home-assistant/core Apache 2.0
Wyoming Voice (Whisper/Piper) Michael Hansen (Synesthesiam) github.com/rhasspy/wyoming-whisper MIT
Wire-Pod (Vector AI) Jason Hendriks & Project Victor github.com/kercre123/wire-pod MIT / AGPLv3
Proxmox VE Proxmox Server Solutions GmbH git.proxmox.com AGPLv3
VictoriaMetrics Aliaksandr Valialkin & Team github.com/VictoriaMetrics/VictoriaMetrics Apache 2.0
Caddy Server Matthew Holt & Caddy Community github.com/caddyserver/caddy Apache 2.0