OM1: Modular by Design, Hardware-Agnostic by Default (Build 2026 DEM306)

Prachi Sethi introduces OM1, a modular orchestration layer for robotics that aims to decouple robot hardware from higher-level cognition. The session walks through OM1’s architecture, shows how capabilities can be swapped in and out, and demonstrates LLM-driven decision-making with both a physical robot and a cloud simulator.

Overview

What OM1 is trying to solve

Robots are often tied to a specific stack: hardware, interfaces, and software components that don’t transfer well to other robots. OM1 is presented as a modular, hardware-agnostic orchestration layer intended to bridge different robots to a shared “cognitive infrastructure” so the same higher-level capabilities can be reused across platforms.

Architecture overview

The session covers OM1’s architecture at a high level, including:

LLM-based robot cognition and decision-making

A core theme is using LLM-based cognition for robot behavior and decision-making. The session highlights:

Demos: physical robot and simulator

The session includes demonstrations:

Performance and implementation notes

The session calls out a migration from Python to Go, motivated by:

Extensibility: “app store” concept

The talk introduces an “app store” concept for robot applications, framing OM1 as a platform where robot capabilities can be packaged and distributed as modular apps.

Resources