We currently consider two frames: odom and vehicle_ned. That won't be enough to handle everything we want to do this year.
We will now have four frames: world, odom, mission, and vehicle.
vehicle is whatever we define to be the vehicle frame (likely positioned at IMU, x forward, y right)
odom is the frame in which state estimation reports and motion planning runs.
world is the frame in which persistent object locations (buckets!) are tracked.
mission is the frame that we "retare" when convenient so we can work in a specific coordinate frame.
The frames will be related as follows.
State estimation publishes the transform between odom and vehicle, which starts as identity.
Other future localization that tells us our position relative to a fixed object publishes the transform between world and odom to correct for state estimation drift.
We define the transform between world and mission as convenient, i.e. as the transform from world to vehicle to retare to our current transform.
micahreich
commented
1 year ago 
We currently consider two frames:
odomandvehicle_ned. That won't be enough to handle everything we want to do this year.We will now have four frames:
world,odom,mission, andvehicle.vehicleis whatever we define to be the vehicle frame (likely positioned at IMU, x forward, y right)odomis the frame in which state estimation reports and motion planning runs.worldis the frame in which persistent object locations (buckets!) are tracked.missionis the frame that we "retare" when convenient so we can work in a specific coordinate frame.The frames will be related as follows.
odomandvehicle, which starts as identity.worldandodomto correct for state estimation drift.worldandmissionas convenient, i.e. as the transform fromworldtovehicleto retare to our current transform.