Reptile Context Protocol (RCP) 🤖

A hardware-independent protocol that gives LLMs the ability to control any robot using natural language.

🎯 Motivation

The multi-disciplinary nature of robotics makes it nearly impossible to create versatile and low cost AI solutions. Different robots require different control systems, creating massive barriers to AI integration, but we realized that most AI robotics integrations adopt the same methodologies, so why not standardize this?

RCP solves this by using natural language - which has denser meaning and better semantics than traditional control interfaces. Inspired by MCP, RCP enables AI agents to directly control robots regardless of hardware configuration. The AI agent handles the rest, namely planning, execution, and adaptation.

Workflow

User Command → Planning Agent (LLM) → Caching Layer (Vector DB) → Execution Agent (SLM) → RCP Server → Physical Hardware

How it works

The client (AI agent app) is exposed to details and tools to control the robot with, and is responsible for the human interaction layer.
The middleware defines abstractions and communications with the robot
The robot exposes it

We can add any data sources, model providers and spend minimal time on integrating inputs and outputs of different modalities. The AI agent is responsible for decoupling low level complexity from the high level and abstract concepts close to the user experience.

About Reptile

Born in the Digital Makerspace at Morrissette's Entrepeneurship Building, Reptile (because it's green lol) is our first step to realizing our vision of scalable and low cost robotics using AI agents as the center. Reptile utilizes in-context learning from our 2 nights of experimentation to learn generalizable movements, with few shot prompts stored as "instruction manuals".