inFuse

h1. 01/16

*Participants:* 

* From MAG: Vincent Bissonnette, Clément Bazerque, Raphaël (?)

* From LAAS: Andrea, Ellon, Quentin

*Goal of the meeting:* As we are going to test our algorithm chains in MORSE, we need to coordinate our effort on the simulator and collaborate where we can. As Clément already started working full-time on MORSE, we went there to see what was already available and how we can catch-up and improve.

h2. MORSE at MAG

MAG's goal at the moment is to produce a full working chain integrated into MORSE (and if possible, with our help, directly on our rovers), *by the end of February*. The first version of the processing chains they want to integrate is the following:

!/home/quentin/Desktop/Chain1.png!

<pre>

Stereovision => Disparity computation => DEM building => Navigation Map => Path planning => Motion control

</pre>

All the algorithms come from CNES and need integration inside MORSE. The following sensor/actuator flows are already implemented:

* Lidar

* Stereobench

* Pan-tilt unit

* Robot control (did not ask which one, but from Simon's report it is a RMP440)

At the moment, the interface between MORSE and the algorithms is made via YARP, but they are working on an interface with ROS. The orchestration of scenarios is made through a simple state-machine written in Python. The scenario implemented so far is "Acquire a scan and move forward for a given distance". No localization is performed in MORSE.The pose of the robot is directly given through MORSE and does not come from a MAG algorithm. The machine has then two states:

* Move pan-tilt

* Go to point

*Concerning datatypes:* Right now, they are not using ASN.1 compliant datatypes. Vincent told us that the ESROCOS ASN.1 compiler exists but the did not interface anything with the C struct produced. He proposes that we use ROS datatypes for now and see for interfacing properly afterwards. We agreed on this.

h2. How LAAS can contribute

*First thing is to provide MAG with accurate simulations of our rovers, in order to test their algorithms and ease integration directly on the rover. They require:

** Sensor streams:

*** Velodyne

*** Stereovision

*** Sick LDMRS

** Transform tree of the rovers

** Accurate control flow

*Second element would be to provide a localization module to integrate in the chain, in order to have the full chains coming from InFuse algorithms => We propose to use the Visual Odometry from Andrea

*Third element is to provide a basic version for the DPM. We would then have the following chain of processing:

<pre>

Stereovision => Disparity => DEM

||=> Visual Odometry ======= ^

</pre>