Skip to content

HackMIT Quickstart

Jump to bottom
ruthwikdasyam edited this page Sep 16, 2026 · 4 revisions

Dimensional (dimOS) - Workshop Quickstart

Workshop run at HackMIT.

This is the install-along-with-us guide: how to get dimos running in simulation, add agentic control, and (if you bring a robot) deploy to real Unitree Go2 hardware.

What is Dimensional?

Dimensional (dimos) is an open-source, agent-native operating system for physical space. It lets you command robots - quadrupeds, humanoids, arms, drones - in natural language, and build multi-agent systems that plug directly into a robot's perception (camera, lidar), spatial memory, and control loops. It works across most humanoids, quadrupeds, and arms on the market (Unitree Go2/G1, Xarm, AgileX Piper, drones via MAVLink/DJI, and more).

Capabilities you'll see today

  • Agentic navigation
  • Spatial memory
  • Agentic manipulation
  • Teleoperation

Example use cases: security patrol, construction-site monitoring.

1. Install

Follow the OS-specific guide:

You only need the system-dependency install steps on that page - not the rest of the docs site. Everything else you'll actually run is spelled out below.

Then clone and set up the project:

git clone https://github.com/dimensionalOS/dimos.git
cd dimos
uv sync --all-extra

uv sync installs dependencies and drops you into the project's virtual environment automatically. If you're not in the venv afterward, activate it manually:

source .venv/bin/activate

2. Run Go2 in simulation

dimos --simulation run unitree-go2

This launches:

  • Rerun - visualization tool (camera, lidar, map)
  • Mujoco - physics simulator Click Keyboard Teleop (bottom-right of the rerun window), then drive the Go2: WASD - Drive, QE - Strafing

3. Add agentic control (simulation)

Requires an OpenAI API key.

OPENAI_API_KEY=<your-api-key> dimos --simulation run unitree-go2-agentic

Same simulation as Step 2, but with an agent layer on top. This time, open humancli and give the agent a natural-language instruction (e.g. "explore the room," or ask it a question) instead of driving manually with keys.

You can also replay the sf_office example to see how a session's data is recorded and saved.

4. Run on real Go2 hardware

Set the robot's IP, then drop the --simulation flag:

ROBOT_IP=<your-robot-ip> dimos run unitree-go2-agentic

Let the robot explore. Image, lidar, and map data build up live in Rerun.

Also worth having open, to see what's happening under the hood:

  • agentspy - monitors agent messages and tool calls
  • lcmspy - monitors LCM messages (robot's internal comms) in real time
  • humancli - interactive terminal for sending messages/prompts to the running agent

5. Manipulation: OpenYAM Hardware

dimos can setup can0
dimos run keyboard-teleop-openyam

Shows a viser view of the arm and obstacles (and/or Rerun for the camera feed). and Teleop with Keyboard.

Tool glossary

  • Rerun - visualization tool for camera, lidar, and map data
  • Mujoco - physics/robot simulator
  • viser - 3D viewer for arm + obstacle visualization
  • humancli - interactive terminal for sending natural-language prompts to the agent
  • agentspy - monitors agent messages and tool calls
  • lcmspy - monitors LCM messages (robot's internal comms) in real time

Resources

Clone this wiki locally