Embrained

Train and deploy neural navigation agents on low-cost robots — locally, privately, on your own PC.

Embrained offloads all heavy compute from the robot to your local GPU over WiFi. You collect real-world data by driving the robot around your home, train vision encoders and navigation policies on your machine, then deploy them back to the robot for autonomous goal-directed navigation. No cloud. No expensive onboard hardware.

Quick Start

1. Install

git clone https://github.com/Embrained/embrained-app.git
cd embrained-app

Windows: setup.bat · Mac/Linux: ./setup.sh

This creates an isolated virtual environment with Python 3.12, PyTorch, and Node.js.

2. Launch

Windows: start.bat · Mac/Linux: ./start.sh

Open http://localhost:8080 to access the dashboard for teleoperation, live telemetry, and autonomy control.

3. Train

Activate your virtual environment first (venv\Scripts\activate.bat on Windows, source venv/bin/activate on Mac/Linux), then run from the project root:

# 1. Consolidate raw transition data into a training dataset
python cognitive-engine/scripts/prepare_dataset.py

# 2. Train the VQ-VAE encoder (compresses camera frames into discrete latents)
python cognitive-engine/backend/training/train_vqvae.py

# 3. Train the CQL navigation policy (maps latents to motor commands)
python cognitive-engine/backend/training/train_discrete_fixed_goal_cql.py

4. Deploy

1. Power on your Plexus robot and verify its WiFi connection in the app.
2. Open the Autonomy Panel, select your trained models, and click Start Autonomy.

The robot navigates to goal locations in real time, running entirely on your local hardware.

---

Optional Setup

CUDA (GPU acceleration): The default install is CPU-only. For faster training with an NVIDIA GPU:

pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu121

Simulation (PyBullet): Requires a C/C++ compiler (MSVC Build Tools on Windows):

pip install -r requirements-simulation.txt

---

Community

Share your trained weights, datasets, and demo videos on our Discord. Contributed model weights feed into a federated Model Soup — enabling generalization across diverse home environments without sharing raw data.

License

• Software — GPLv3. Open to audit, modify, and contribute.
• Your locally-trained weights — 100% yours, no restrictions.
• Pre-trained / aggregated weights distributed by Embrained — governed by the Embrained Open-Weights EULA (non-commercial use permitted; commercial use requires a license). See LICENSE for full terms.