CrowdScout

[1306scouting]

Welcome to the Future of Scouting

Team 1306 would like to invite teams at Championships to join CrowdScout, FRC Team 1306's new scouting initiative.

CrowdScout is Team 1306's new initiative for collaborative scouting. We can cut down the amount of time participants spend scouting at an event by a huge amount by working together and sharing data, freeing up more time to enjoy the tournament and participate in the myriad of other things that occur at FIRST Championship. If we have 150 matches, and 102 teams with 6 scouters per team, collectively we will spend 143 days actively scouting, discounting time waiting for field reset and so forth. If half the teams on Curie worked together with CrowdScout, they would spend around 2 days scouting. When we ran CrowdScout at the Wisconsin Regional, we worked with Piratech (4804) and both teams cut their scouting loads in half, allowing us to free up scouters to participate in other parts of the tournament.

We do this by having fewer scouters taking redundant data on each match. With our current plan, we have two scouters focusing on each robot per match, though if everyone involved would rather have more for either more redundant data or more data points CrowdScout is flexible enough to accommodate. Furthermore, CrowdScout could intake data from many sources - while 1306 uses a laminated paper sheet system because we prefer it, we could easily integrate data from teams with more technologically advanced scouting systems. (If you're curious about integrating your data, PM me or reply to this thread.)

About Divisions

1306 is in the Curie Division, and thus will have a CrowdScout program running in Curie. If there is interest in having CrowdScout in other divisions, we can run it for other divisions too. Alternately, if there is a team in another division who would be interested in coordinating CrowdScout, we can help them through the process.

Why CrowdScout?

CrowdScout decreases the amount of time you and your team spend scouting in the stands. The more teams participate, the less time each team has to spend scouting; if half a division participates, the time scouting for each team will be less than two hours, as compared to 33 hours without CrowdScout.

1306's Scouting System

1306 uses a laminated scouting sheet with wet erase markers for the scouters. (There are many reasons behind our preference for such a system over a tablet/computer/phone based one, but we see why other teams use them.) We then have a runner take these sheets to our scouting station, where we scan them into a computer and then run automated recognition software, exporting the data to a .json which we can share with others and run through our own processing algorithms to generate our alliance selection lists. (Our scouting sheet from the Wisconsin Regional is attached.)

If you have any questions, feel free to ask me via PM, our subreddit, or by posting below.

[Grim Tuesday]

I'm interested in how your algorithm that spits out your ideal alliance selection works.

[1306scouting]

Originally Posted by Grim Tuesday

I'm interested in how your algorithm that spits out your ideal alliance selection works.

Our algorithm doesn't give us our final list, as there are factors which are not shown in quantitative data.

We have a python script which we give weightings for various robot abilities (for instance, to look for a shooting robot, we give a higher weight on shooting points) that outputs a sorted list of teams based on those factors. The scouting team then goes through this list and finds alliances that would work well and complement us. (Disclaimer: 1306 does not have a highly competitive robot this year, and thus we created lists as an exercise but not with the intentions of using the lists.)

For instance, if we were targeting a full-court shooter based alliance, we would create a list based on shooting score/ability. We would then take the teams on that list which are full court shooters, and those would be our first picks. Since we would likely be playing anti-defense, we would then find a complementary bot, in this example a floor-loading bot, by cross-referencing our pit scouting data of floor-loading robots and a list of shooting potential mixed with climbing score. That would form our alliance list for a full-court shooter based alliance, and the alliance selection representative would then work based off their knowledge of the field, input from the drive team, and the lists to pick the alliance.

[davidfv]

I was wondering if teams that participate would just be able to get the raw data that they could use to run their own weighting system on... A spreadsheet with all the teams final stats?

[AlexanderKahn]

This system seems very good, but small things to change would be adding a third column to the round number, obviously. When all data is inputed, are you able to transfer all of the data through a single document for other teams to use? or only use lists? And what is your system for pit scouting?

[1306scouting]

Originally Posted by davidfv

I was wondering if teams that participate would just be able to get the raw data that they could use to run their own weighting system on... A spreadsheet with all the teams final stats?

We will be releasing several files: the raw data in a .json, a more compiled spreadsheet (final stats), and lists based on points by shooting and climbing. If a team wants any other data, they should feel free to ask us; the aforementioned 4 data sets are just the ones we have considered.

I can try to get examples of these files (from the Wisconsin Regional) tomorrow if there is interest.

Originally Posted by AlexanderKahn

This system seems very good, but small things to change would be adding a third column to the round number, obviously. [snip] And what is your system for pit scouting?

We will be revisiting the scouting sheet shortly to make changes like adding a third digit to the round number and changes like that; nonetheless, thank you for reminding us.

Pit scouting, for us, is something that needs to be done by each team to find which robots would accommodate them; as such, we don't currently have a plan for sharing pit scouting data. I would suggest starting with the spreadsheets. (I have heard these are particularly good.)

[depth_Finder]

I would be very interested in seeing those files. Especially the raw data and the spreadsheet.

[1306scouting]

Originally Posted by depth_Finder

I would be very interested in seeing those files. Especially the raw data and the spreadsheet.

While I'm still working on obtaining a copy of the spreadsheet, I have attached a formatted copy of the raw data from the Wisconsin Regional. The only difference between the attached file and the one that we export is that our exported one is a single line. (The attached file is a .txt since CD doesn't allow JSON attachments. The JSON can be found here, too.)

We are putting the finishing touches on a slightly different format for Championships, which will have fewer extraneous fields in the data; essentially, the automated mark recognition software we use is build for teachers making multiple-choice tests in LaTeX, so we're building a custom solution that's more stable and actually purpose-built. Once it's done, we will have it open-sourced at our GitHub for anyone interested in seeing it or contributing it.

As always, if anyone has any questions, feel free to reply in the thread or PM me.

[Kathmandu]

Team 3374 is in the Curie Division and would be interested in the Crowd Source scouting. We have about 14 team members who are going to scout. Previously we've used a smartphone app, but haven't been very successful with weighting the results. We'd like to learn from how you transfer the data into something useful. To be honest - don't know what json is.

[1306scouting]

Originally Posted by Kathmandu

Team 3374 is in the Curie Division and would be interested in the Crowd Source scouting. We have about 14 team members who are going to scout. Previously we've used a smartphone app, but haven't been very successful with weighting the results. We'd like to learn from how you transfer the data into something useful. To be honest - don't know what json is.

JSON is a data storage format, much like a spreadsheet, which is (somewhat) human-readable but mainly built for computers to use internally. It's just a generally better way to store data than a proprietary data format, since there are lots of libraries for interacting with JSON in almost any programming language you could want.

To move from JSON to a more useful data format, we use two approaches. The first approach is that we create a spreadsheet (I'm afraid I don't have access to one from our regional yet - it seems that spreadsheet was left on our scouting computer, which is already packed.) with the data by averaging the values on a per-team basis. The second approach is that we use some custom python scripts to create lists of teams based on weighting the factors - for instance, we looked for a high-scoring shooter with a reliable climb by factoring in shooting heavily and then putting in a lesser factor of the climbing points.

If anyone would like to join CrowdScout on Curie:

Please PM me your email address so I can add you to the planning email thread.

[Sir Zordor]

I personally do my own scouting/pit scouting no sheets (yes I like to live dangerously ;p) don't get me wrong we have sheets I personally do not use them haha, I just make them anyway, how would I be able to see the raw data, I personally cannot do this as I am Drive coach along with Head Scout, but my team might be able to, but I'd just like to know where I can access the data.

-Preston #Team 1262 Curie

[Sir Zordor]

nevermind found it