Train Your Own AI Robot at Home

Plexus is a low-cost mobile robot engineered specifically for home navigation R&D. Designed from the ground up for vision-based topological navigation under extreme compute and hardware constraints, Plexus gives researchers, hobbyists, and AI enthusiasts a real platform to train and deploy embodied intelligence.

What's in the Box

• The Core: Fully assembled Plexus robot featuring the XIAO ESP32-S3 with camera and antenna
• The Chassis: Acrylic Plate Triplet
• The Drivetrain: 2xTT gearboxes with rubber tires, durable metal motor mounts, and a smooth ball caster
• Sensors & Feedback: Sharp IR distance sensor and tally light LED

The Full AI Pipeline - Runs Locally on Your Laptop

The Plexus software suite delivers a complete, end-to-end pipeline for training embodied AI using a standard mid-range laptop GPU (4–8 GB VRAM). Everything runs locally via the web dashboard and Python CLI:

1. Collect Data: Drive the robot using WASD keys or autonomous random-walk. Every camera frame is paired with its exact motor command, building a rich dataset of visual transitions.
2. Compress Perception: Train a discrete VQ-VAE to compress raw video into a stable latent manifold of tokens - a learned visual vocabulary of your specific environment, without brittle object detection.
3. Train the Policy: Train an offline CQL (Conservative Q-Learning) RL model that learns which discrete motor actions reliably move the robot toward spatial goals using only latent tokens as input.
4. Deploy Autonomy: Load your trained weights into the Embrained Dashboard, hand over control, and watch the AI drive autonomously based on live telemetry.

Privacy-First by Design

Because the entire pipeline runs locally, your raw video feeds and latent data never leave your laptop. Share only your trained neural model weights with the community - enabling collective robot learning while keeping the interior of your home strictly private.