45 posts tagged Robotics. ← Back to all posts
An introduction to the OpenCastor ecosystem — a single ROBOT.md manifest, runtimes in Python and TypeScript, an MCP server, a Claude Code plugin marketplace, a public registry with EU AI Act-aligned compliance endpoints, and the consumer-tier UX I've been focused on lately. Built almost entirely with Claude Code, on a SO-ARM101 arm I call bob.
A two-command install that wires your robot's ROBOT.md into any Claude Code session as structured MCP resources. Also: what RCAN and the Robot Registry Foundation do, and why this plugin belongs in Anthropic's hosted MCP catalog.
External systems need to drive robots too. The obvious move is to put robot-md-mcp behind an HTTPS frontdoor. It works. It's the wrong shape. Here's the right one — a BYOK Claude Agent SDK dispatcher that runs on the robot host.
Between April 17 and April 21, 2026 I shipped ROBOT.md v0.1 through v0.8, an MCP server, a hardware pendant with ESP32-C6 firmware, and the landing page at robotmd.dev — across 364 commits in five repositories. This post documents the resource envelope (tokens, cost, cache-hit rate, commit cadence) and compares the observed working style to Thariq Shihipar's published recommendations for the 1M context window.
Introducing ROBOT.md — a single-file declaration at your robot's root that any agent harness (Claude Code, ChatGPT, Gemini, Ollama, your own) can read at session start. Vendor-neutral. Apache 2.0. Shipping v0.1 today.
What happens when you give Claude Code direct, LoA-gated, RCAN-signed access to a physical robot via the Model Context Protocol.
OpenCastor now exposes a full MCP server with LoA-gated tools. Any AI agent — Claude Code, Codex, Gemini, cron jobs — can command and observe robots through a single interface. Here's why we built it, and how compliance requirements shaped the design.
RCAN v2.1 shipped today. It's a robot protocol that now handles firmware attestation, supply chain transparency, multi-robot fleet authorization, and EU AI Act compliance. Here's the full story — what it does, why it exists, and what scenarios it actually solves.
OpenCastor's new contribute feature turns idle robot compute into a distributed science network. How the open-source runtime layer between robots and AI makes this possible, what RCAN is, and why a robot registry matters.
OpenCastor v2026.3.20.3 adds first-class support for authenticated access to closed AI models — Physical Intelligence π0, HuggingFace gated models, enterprise APIs, and more.
Why we're structuring OpenCastor as a B Corp, how Castor Credits turn idle robot compute into real value, and why the RCAN protocol and Robot Registry need to be owned by foundations — not us.
How a nightly Gemini-powered research pipeline finds the best agent harness configuration for your robot — and why the next version will tune itself differently for every hardware profile.
Today we shipped RCAN v1.5 — addressing 18 security and safety gaps identified in a comprehensive protocol audit. Replay attack prevention, robot identity revocation, training data consent (EU AI Act), delegation chains, and more.
RCAN v1.6 ships the four gaps we deferred from v1.5: federated consent across registries, bandwidth-constrained transport (32-byte ESTOP for LoRa), multi-modal payloads with SHA-256 audit trails, and human identity verification with Level of Assurance.
A detailed walkthrough of the full stack — OpenCastor, RCAN, LLM providers, Protocol 66 safety — behind two Raspberry Pi 5s holding a genuine planning conversation and coordinating on a real task.
Four new hardware detectors and an expanded I2C lookup table land in OpenCastor v2026.3.12.0 — Dynamixel U2D2, RPLidar/YDLIDAR, Raspberry Pi AI Camera, LeRobot SO-ARM101, plus five new I2C sensor types.
Three projects. One complete open stack for robot identity, accountability, and registration. Here's what shipped and why it matters.
castor scan now auto-detects 12+ hardware types. Plus native support for SO-ARM101, Koch arm, ALOHA, and Pollen Robotics Reachy.
Sixteen issues from a fresh Pi reinstall, now fixed. castor scan, doctor, upgrade, stop, gateway hardening, and a proper venv guide.
Two major additions ship today: an EmbeddingInterpreter that gives robots semantic memory using local CLIP embeddings, and first-class support for HLaboratories' ACB v2.0 BLDC motor controller.
Three simultaneous releases ship a complete open-source stack for AI-accountable robotics: rcan-py v0.1.0 on PyPI, @continuonai/rcan-ts v0.1.0 on npm, and OpenCastor v2026.3.6.0 with deep RCAN v1.2 integration.
rcan-py v0.3.0, rcan-ts v0.3.0, OpenCastor v2026.3.8.0, §17 Distributed Registry, and RCAN-Swarm Safety — across four repos and most of the day.
Coordinated releases across four repos, green CI on all of them, a new swarm safety claim for OpenCastor, and the clearest picture yet of what the Robot Registry Foundation actually needs to function.
Every robot incident investigation starts the same way: what did it do, who authorized it, and can you prove it? RCAN is the protocol layer that makes those questions answerable — before the incident happens.
Why AI robots need accountability protocols now, what existing safety standards miss, and how to adopt RCAN before the incident that makes it mandatory.
The automotive industry spent decades standardizing autonomy after the technology was already deployed. Robotics is at the same inflection point. RCAN, OpenCastor, and QuantumLink-Sim are a bet on doing it the other way around.
ISO 10218, OPC UA, IEC 62443, ROS2 — the standards that govern industrial robots today were written before AI was making the decisions. Here's what each covers, where they stop, and where RCAN fits.
Thariq Shihipar from Anthropic published a detailed breakdown of how they built Claude Code. Every design decision maps to OpenCastor — but for a different consequence space. And this week, that consequence space got a lot more political.
AI agents make claims. Those claims drive actions. If you can't verify the chain — what was claimed, when, and under what keys — you can't audit anything. Here's how quantum-link-sim's CommitmentEngine solves that.
Why a robot that improves itself matters more than a fleet of static ones — and the engineering steps to get there.
Today we release OpenCastor v2026.2.19.0 — an open-source framework that connects any AI model to any robot hardware through a single YAML config. 7 providers, tiered brain architecture, Hailo-8 NPU vision, and a $0 starting cost.
ContinuonOS went from a chatbot wrapper to a complete cognitive architecture in 7 phases, all executed by Claude, Clawdbot, and Gemini working together on a Raspberry Pi 5. Here's how the architecture works, what we tested, and the principles that guide a robot brain designed to live with humans.
A detailed breakdown of the parts, Bill of Materials, and grey areas we discovered while building a 6-foot-reach mobile manipulation robot with dual arms, V-slot mast, and EcoFlow power system.
How we built a self-improving training system for ContinuonXR that generates its own learning curriculum, tracks brain state, and prepares data for MambaWave neural networks—all without human intervention.
A deep dive into the RCAN open source project, its Astro-based architecture, and how you can contribute to the future of robot communication.
The technical appendix to the RCAN protocol proposal. Includes JSON Schema for Robot URIs, Protocol Buffer definitions, and a complete handshake flow—proving RCAN is real code, not just a blog post.
The internet has ICANN. IoT has Matter. Robotics has nothing. The RCAN protocol proposes a global addressing, authentication, and governance standard for embodied AI.
2,652 commits, 76 new projects, and the journey from chasing business ideas to just building.
China's dominance in industrial robotics raises fundamental questions about who shapes the values embedded in embodied AI systems that will increasingly act in the physical world.
As embodied AI scales, we need a governance stack for the physical world—a constitutional operating system that encodes safety, law, and social norms into every robot's decision-making framework.
Why the future of robotics isn't a better robot—it's a better Operating System. How ContinuonAI applies UNIX principles to build the operating system for embodied AI.
A technical rationale for the latest Continuon Brain defaults: why they are explicit, how they resolve multi-owner conflicts, and how they preserve safety without degrading swarm throughput.
Testing hypotheses about safety-first autonomy, reproducible robot builds, and what happens when a cognitive architecture lives at the edge instead of the cloud.
Establishing the ContinuonXR robotics stack: Hypothesis validation, Raspberry Pi 5 integration, and the architectural transition to HOPE + CMS.
A modular learning plan that ties theory, coding practice, robotics, and MLOps into a single Nested Learning journey.