Deployment and validation
Pose Engine is deployed as follows:
| Step | Deployment | Notes |
|---|---|---|
| Platform setup | N/A | Process: MSA engineers configure Pose Engine for specific platform |
| Remote service | Cloud | Customer uploads sensor data to Data Portal; Pose Engine generates results; MSA engineers review results before release |
| Platform hardening | N/A | Process: MSA engineers configure Pose Engine for on-prem deployment |
| On-prem (offline) | Docker | Customer runs Docker container on a vehicle computer, workstation, or server (no sensor data is transferred to MSA); Pose Engine generates results |
| Online integration | N/A | Process: MSA engineers work with customer to configure Pose Engine for online operation |
| Online (on-robot) | Library | Pose Engine runs online on the robot (no sensor data is transferred to MSA); Pose Engine generates results; requires site map and bootstrapping |
Platform setup
Platform setup is the process of configuring Pose Engine to ingest and process sensor data from a specific robot platform. This process is performed by MSA engineers and typically takes a few weeks.
After platform setup is complete, Pose Engine can process logs and generate pose results for any instance of that robot platform.
Remote service
Remote service (Pose-as-a-Service) is the cloud-hosted deployment of Pose Engine. Sensor data is uploaded to the Data Portal and processed by MSA with human review.
- Customer uploads sensor data (from an instance of a robot platform that has been setup) to the Data Portal
- Pose Engine generates a pose output
- MSA engineers review the pose output
- The pose output is made available for download from the Data Portal
Platform hardening
Platform hardening is the process of ensuring Pose Engine can generate pose outputs without human review. This process is managed by MSA engineers, requires sensor data collected from different instances of the same platform, and typically takes a few months.
After platform hardening is complete, Pose Engine can process logs and generate results without human review.
On-prem (offline), requires hardening
On-prem deployment requires platform hardening.
On-prem deployment runs Pose Engine in a Docker container on a vehicle computer, workstation, or server. No sensor data is transferred to MSA.
- MSA provides a Docker image configured for a specific platform
- The Docker image is run in a container on a local computer or in the cloud
- Sensor data (from an instance of a robot platform that has been setup) is referenced when the Docker container is launched
- Pose Engine generates a pose output
- The pose output is made available in a target directory
Online integration
Online integration is the process of integrating Pose Engine into a customer’s autonomy stack. This process is a joint activity with MSA and customer engineers and typically takes a few months.
After online integration is complete, Pose Engine can run online on the robot.
Online (on-robot), requires integration
Online deployment requires platform hardening, site mapping, and an online integration effort.
Online deployment uses MSA-provided software that runs on a robot belonging to a specific platform:
- MSA provides a software library configured for a specific platform
- Pose Engine software runs online on the robot
- Sensor data is provided through a pub/sub API
- Pose Engine generates odometry and pose outputs
Pose Engine online deployments require:
- CPU-only; no GPU required; 3–4 cores on Jetson Orin or equivalent
- A site map built and validated before online deployment
- Bootstrapping hint for initial localization to the map