Case study
How Overland AI calibrates sensors on autonomous ground vehicles used by the US Military
Problem
- Sensor calibration, ~20 sensors / vehicle
- Needs to happen in any outdoor location
- Calibration process must be fast and reliable
- Sensor data cannot be offloaded from the vehicle
Solution
- Calibrates any number and combination of sensors
- Works anywhere: targetless, no checkerboards
- Runs in a Docker container on the vehicle
Company
- Name
- Overland AI
- Domain
- Off-road autonomous ground robotics
- Platform
- Ultra, fully autonomous tactical ground vehicle
- Fleet
- Scaling to hundreds
Featured engineer
Director of Integration and Delivery, Overland AI
Role: Leads Ultra software deployment, compute and sensor platform, and calibration integration
Backstory: UW postdoc advisor was Byron Boots, Overland CEO; previously tackled calibration on agile drones at CMU, and underwater robots at Cornell
We don't have to worry about a problem that plagues every robot. Knowing the calibration is correct allows us to focus on the broader autonomy mission.
Previous solution
Overland’s previous calibration process was manual and target-based: the process required raising the vehicle on dollies to rotate it in place. It needed several engineers, needed dedicated space, and couldn’t scale into field deployments. The calibration result also wasn’t repeatable. Calibration happened infrequently with failures that were hard to diagnose.
Limitations of Previous Solution
- Required multiple engineers and took almost a full day per calibration
- Required dollies, targets, checkerboards, and a specific known environment
- Generated an unreliable result that was operator-dependent
The Calibration Anywhere solution