How to run autonomous inspections with a Unitree Go2
Which Go2 models support autonomous inspection, what the DIY ROS 2 stack involves, and how to go from unboxing to scheduled inspection missions in the browser.
OLO Robotics Team · · 7 min read
The Unitree Go2 made capable quadrupeds affordable, and that changed who buys them: alongside research labs, it is now facilities teams, inspection contractors, and operations groups who want a robot walking their rounds. The hardware is ready for that job. What the box does not include is the software layer that turns a remote-controlled robot dog into an autonomous inspector following a repeatable route and recording what it finds. This guide covers what you need: the right Go2 variant, the software options, and the practical path to a first autonomous mission.
The short answer
To run autonomous inspections with a Unitree Go2 you need three things: a Go2 variant with SDK access (the developer and enterprise models, not the consumer ones), a ROS 2 connection to the robot, and mission software that handles routes, data capture, and scheduling. You can assemble that last layer yourself from open-source ROS 2 components, or use a platform like OLO that provides it out of the box from the browser.
Which Go2 models can run autonomous inspections?
Unitree segments the Go2 range by how much access you get to the robot. The consumer models are driven from a phone app and do not expose the interfaces autonomous work needs. For inspection you want a variant with SDK access and ROS 2 support — Unitree publishes official SDK and ROS 2 packages (unitree_sdk2 and unitree_ros2) that cover the Go2 developer and enterprise lines.
- Go2 Edu — the usual choice for development and research: full SDK and ROS 2 support for secondary development.
- Go2 enterprise variants — configured around inspection payloads such as gas detection and pan-tilt-zoom camera modules.
- Go2-W — the wheeled variant, suited to mixed-terrain patrols where distance matters.
Option 1: build your own ROS 2 inspection stack
The open-source route is well trodden and genuinely good — if you have robotics engineering capacity. A typical DIY stack combines the unitree_ros2 driver with Nav2 for navigation, a SLAM package such as RTAB-Map for mapping, and custom nodes for detection and reporting. Public projects show what is achievable: community stacks pair the Go2 with LiDAR odometry, terrain analysis, GPU-accelerated planners, and even automated report generation.
The honest costs are time and maintenance. You are integrating, tuning, and owning a distributed system: drivers break with firmware updates, SLAM needs per-site tuning, and the "boring" parts — scheduling, data management, dashboards your operations team can actually use — are usually the last to be built and the first to be missed. For a research group this is the point; for an operations team it is a detour.
Option 2: use an inspection platform
The platform route treats the Go2 as what it is to your business: a data-collection vehicle. OLO supports Unitree natively — an established integration, not a project — with the robot-side Appliance joining the Go2's ROS 2 environment and linking it to the browser Portal: visualisation, teleoperation, mission configuration, and a scripting playground, with nothing installed on your team's machines. The inspection workflow is the same five steps on every site:
- Simulate — import a 3D model of your environment and trial the inspection virtually before the robot moves.
- Map — generate a live SLAM map of the real site with the OLO visualiser.
- Configure — place waypoints, set the action at each stop (photo, thermal capture, sensor reading), and lay out the dashboard.
- Deploy — schedule the mission; the Go2 walks the route autonomously.
- Analyse — review live feeds and the complete, timestamped ROSBag recording of every run.
Most sites run their first autonomous route within half a day. Enterprise plans start at £175 per seat per month, and a free personal plan lets you evaluate the workflow in simulation before the robot is even delivered.
DIY stack vs platform: an honest comparison
| DIY ROS 2 stack | Inspection platform (OLO) | |
|---|---|---|
| Best for | Research groups, teams with ROS engineers, novel requirements | Operations teams that need inspections running, not a robotics project |
| Time to first mission | Weeks to months, depending on experience | Typically half a day on site |
| Upfront cost | Free software; significant engineering time | From £175/seat/month; minimal engineering time |
| Maintenance | You own drivers, tuning, and breakage | Handled by the platform |
| Flexibility | Unlimited — it is your code | Portal workflow, plus typed Python/TypeScript SDKs when you need custom logic |
Payloads and the data you get back
Inspection value lives in the payload data. Anything that publishes over ROS 2 can be captured at each waypoint — pan-tilt-zoom cameras, thermal imagers, and gas sensors are the common set on quadrupeds. Because the route and camera positions repeat exactly, readings are comparable between runs: the same gauge from the same angle, every time, with everything timestamped in the run's ROSBag recording.
Getting started checklist
- Choose a Go2 variant with SDK and ROS 2 access (Edu or enterprise).
- Confirm the payloads your inspections need (visual, thermal, gas).
- Get a 3D model of your site — CAD or a scan — for simulation.
- Decide DIY vs platform. If platform, tell us the model and use case and we will show your inspection running in simulation first.
For the broader picture of what quadruped mission software should do, see our guide to robot dog inspection software.