Milo: Open-source robotic guide dog costs $2k, navigates indoors and outdoors
A fully autonomous robot guide dog for $2k, built on a Unitree Go2.
Researchers from academia and industry have open-sourced Milo, a fully autonomous robotic guide dog built on a modified Unitree Go2 quadruped platform. The system costs roughly $2,000—dramatically undercutting the $50,000 price tag and ongoing maintenance of a real guide dog. Milo requires no prior 3D maps of its environment; all computation runs onboard, making it suitable for both indoor and outdoor navigation.
Its perception stack combines voxel mapping with floor, obstacle, and pedestrian detection. The navigation stack relies on an obstacle-avoidance policy trained in a custom bird's-eye-view simulator. In real-world obstacle courses, Milo demonstrated smoother navigation and fewer handler collisions compared to a costmap-based baseline. The team has published the full hardware instructions and software stack to maximize accessibility for blind and low-vision users.
- Milo costs ~$2k, compared to ~$50k for a trained guide dog plus ongoing expenses.
- Fully autonomous: no prior environment scans or external compute needed, works indoors and outdoors.
- Built on a Unitree Go2 robot with onboard sensors and handle; entire stack open-sourced.
Why It Matters
Milo could drastically lower the cost and waiting time for guide dogs, increasing mobility for blind and low-vision individuals.