search

strands-project / trajectory_behaviours

Apply qualitative spatial temporal relationships between detected trajectories and static objects, per ROI.
2 stars 8 forks source link

Schedule visits to under-observered areas. #6

Open hawesie opened 9 years ago

hawesie commented 9 years ago

For @eris-ch's temporal novelty scores, it would be good to schedule the robot to visit under-observed area / time combinations . If you can write code to print out these times, it will be trivial to schedule robot actions to do this.

PDuckworth commented 9 years ago

@hawesie , what do you mean by "print out"? I have the scores as a dictionary (per region, per hour) in my novelty service. They are updated in the offline learning stage (when charging).

PS I moved time_analysis under trajectory_behaviours repo. It made sense.

hawesie commented 9 years ago

Maybe I should just look at that novelty service and see what I can do.

You will need to pull request your changes into this repo so it's in the central strands account.

PDuckworth commented 9 years ago

@hawesie @eris-ch See attached for the information we create about the monitored regions, at different times. But currently this is based on monitored nav failures. Any idea about the availability of all the robot pose data in mongodb?

roi_knowledge

hawesie commented 9 years ago

We will start logging it when we start on the 1st floor of CS, from tomorrow onwards. More information on that later.

PDuckworth commented 9 years ago

Awesome thanks. We just need to know which mongodb msg store to look in, and also maybe think about a sensible sampling rate. one a second?

hawesie commented 9 years ago

If you have a running system (simulated or robot) you can run this:

rosrun mongodb_log mongodb_log.py /robot_pose

This is pretty much what we'll use to log the robot pose, so this will give you the information you need.

@lucasb-eyer is this what you are doing for the node-based logger?

lucasb-eyer commented 9 years ago

Essentially, yes. But it's being merged together with the UBD detection in my case and bayes track in @cdondrup's case, so neither of them are continuous in that sense.

hawesie commented 9 years ago

Ok, then we need to run a separate logger for robot_pose for @PDuckworth. This will be done as above, using mongodb_log.

cdondrup commented 9 years ago

Aren't we logging tf any way?

hawesie commented 9 years ago

@cdondrup I don't know. We need to find a joint resolution and launch file / tmux entry for https://github.com/strands-project/aaf_deployment/issues/17 I guess.

PDuckworth commented 9 years ago

Thanks @hawesie. How do I query the data stored by mongodb_log? Is it just in a normal message_store? If so, which?

I currently still query monitored_nav_events message score for my "Region Knowledge".

PDuckworth commented 9 years ago

@hawesie This is resolved in https://github.com/strands-project/trajectory_behaviours/pull/14

I upload the day's region knowledge (which regions the robot has been looking at during the day) to region_knowledge message store. Where it is queried the next day and incrementally updated. (Hopefully before you remove roslog :)

This can also be queried by anyone looking to send the robot to "under-observed areas".