Overview

Konnex subnets compete on control and perception policies for physical tasks, not on unconstrained chat. The documentation in this section describes VLA (vision–language–action) and related models that teams integrate behind subnet-specific interfaces.

How models connect to the network

1. Task — A user or operator issues an instruction (natural language or schema-defined) for a workload class.

2. Miners — Propose trajectories, policies, or reconstructions; attach whatever the subnet API requires.

3. Validators — Run subnet scoring (simulation, sensor checks, ground-truth comparison) and emit results that end up in Proof-of-Physical-Work records.

Pages in this section

Manipulation & VLA

• Roboarm VLA subnet — workload page

• OpenVLA

• OpenVLA · OFT

• PI-0

• PI-0.5

Aerial (drone navigation)

• Drone navigation subnet — workload page

• OpenFly & OpenFly-Agent — aerial vision–language navigation

SLAM 3D map

• SLAM 3D map subnet — workload page

• ORB-SLAM3 & RTAB-Map — visual SLAM and dense mapping stacks

Infrastructure

• AI verifier (scoring)

• AI fetch interface

For broader network context, see the decentralized AI ecosystem.