Getting Ready for January

Last week we got a number of things done:

1) We talked about what we had accomplished up through the regional tournament and each came up with "two stars and a wish" - that is, two things we thought we did well this season, and one thing we'd like to work on some more.  Between now and the provincial tournament we're going to try to make more progress on the "wishes".

2) We learned how to make and edit 3D models using a program called Sculptris.  Remember, if we're going to train a computer to recognize particular animals, we need to give it lots of examples to learn from.  Those examples can't all be the same.  They have to have enough variety that the computer can learn to tell the animal we're looking for from other animals and non-animals.  We bought a bag of toy cows to train the computer on what cows look like, but the cows in the bag were all identical.  That won't work because the computer will think all cows are exactly the same.  So instead we got a 3D model of a wildebeest and each took turns putting it in different poses to get some variety.  Since last week we printed out all the different animals on our 3D printer, as you can see below.

Ten Wildebeest designed by the Comet Warriors

With wildebeest models in a dozen different sizes and poses we can now do a better job teaching the computer what to expect.

3) Running the robot.  At the tournament, Olivia (also le mot de la semaine), Lily, and Macy were responsible for running the robot.  Last week they worked to teach others how to line up and run the robot, change attachments, switch missions, and so on.  Hopefully, by the end of the season everyone will know how to do this.

4) Planned our next fundraiser.  We plan to sell reindeer hot chocolate kits at the school Christmas party.  This was very successful last year so we're going to try it again.  We'll spend some time this week on putting the kits together.

Homework for this week:

1) Take a look at the Grevy's Zebra Website  This site explains how the Grevy's Zebra Trust did a census of zebras in Kenya.  It involved a lot of driving around with cameras.  A group at Princeton then used computers to try to tell one zebra from another based on the pattern of stripes on their hip.  The woman who runs the Grevy's Zebra Trust, Belinda Mackey, is one of the people we shared our research with.  The video below is a talk she gave in San Francisco in October.  It's half an hour long.  It's worth watching, but if you don't have time for the whole thing watch at least the last six minutes - starting at 25:00.  It gives great examples of how to get people involved in conservation.  It's not as simple as showing up with a new technology.

Belinda Mackey's talk on Zebra Conservation

Lastly, some good news from one of our experts!  When we had our yard sale in October, John Abrams stopped by and was interested in what we were doing.  He has a friend, Tim Lillicrap, who works for Google Deep Mind in London, England.  Well John put us in touch with Tim, and Tim has offered to come visit us on Dec. 22, the last meeting of the year.  Tim works on some of the coolest machine learning stuff in the world (like teaching robots to dream) and is interested to see what we're up to.  Maybe he can give us some ideas on how to make it better.

See you Thursday!

Congratulations, Comet Warriors!

The Comet Warriors competed at the Kingston Regional FLL tournament and did very well, coming away with the Robot Performance Award for the highest scoring robot, as well as the Champions Award for the best performance across robot performance, research, and core values.  Everybody on the team, from the oldest to the youngest, made important contributions to our success on Saturday, and you all performed extremely well.  It was great fun to watch. Congratulations!

What this means is that it's off to Oshawa in January for the provincial tournament, where the other participants will be the top 10 percent of teams from downtown Toronto all the way east to the Quebec border.  That's a lot of area and a lot of really good teams.  We're going to need to do a lot more work between now and January if we want to represent ECC and Kingston well.

For the next week, though, we can relax a bit.  So there's no homework this week.

Picking Up Some Hardware

Got A Few Problems Sorted Out

Trying to Figure Out What's Going Wrong

10 Days to Go

Good progress over the past week, Comets!  Since our last post we have:

1) Put together a first version of our presentation.
2) Gotten two more robot missions to work: getting food out of the fridge, and feeding some of the animals.
3) Gotten movement detection working for our project.  That is, we got the computer to look at videos of the cows and identify them by the fact that they're moving.

For those looking for the word of the week, since we got the fridge mission to work, the word for this week is Fridge.

Some things we need to work on for homework for this week, and over the next ten days, are:
1) Everybody needs to learn their lines for our presentation.  You've all received a copy of the script either on paper or in email, so make sure you learn your part.
2) Everybody needs to think back on what they've personally done so far this season.  A lot of this is covered right here on the blog, so make sure you re-read it and know what your responsibilities were.
3) Some of you got individual assignments - writing emails, doing research, making props, etc. - that you need to do at home.

Also, we're going to try to meet every day at lunch between now and the tournament on Nov. 19th.  We'll also be having more evening and weekend meetings, which you should have received emails on.  This past weekend we had an impromptu "plaid day" to work on robot programming.  Check it out:

I was just joking about the word being Fridge.  It's actually "pez" because the fridge works more like a pez dispenser, and so does our attachment.

Less than three weeks to go

Good job last week, Warriors!  Especially to those participating in the presentation at the Toronto Audio Visual and Electronics Show on separating silicon from silicon carbide.  The presentation went very well and I think everyone found it entertaining and informative.

We made more progress on all fronts including:

a) The teamwork and communication show during the "build what we built" exercise was very good.  I especially liked the "E-C-O-L-E C-O-M-E-T-S Go Comets!"

b) Everybody learned to fly the drone.

c) Everybody learned to scan things in 3D

d) Nick, Alyssa, and Lais learned to convert a collection of pictures into a 3D model.

e) Tara and Macy learned to operate the infrared camera.

c) Lily and Olivia learned to use the gyro sensor.

In addition, after the Electronics show Tara, Olivia, Helen and I went to Lucknow, Ontario to see if we could get some drone pictures of cows.  Tara and Olivia's aunt and uncle (Paul and Christine - words of the week) raise cattle.  They let the girls fly the drone above the cows to see what we could learn about counting them.  They got a bunch of pictures which they stitched together into the image below, and if you zoom in you can see the individual cows in the closeup below that.

Stitched Images

Aerial Closeup

They also tried taking pictures with the infrared camera.  They attached the camera to the bottom of the drone and got one good picture of the car as the drone went up.  Unfortunately the camera fell off the drone almost immediately, but we were able to find it.  They took pictures from the ground after that, and are looking for a better way to attach it.  The pictures look promising, though.  Below are the pictures of the car from above, with the yellow area being the warm hood over the engine.  Also, there's a normal picture of the cows at the fence with another picture taken with the infrared camera showing that the cows are warm.

Car with a Warm Hood

Cows at the Fence

Infrared Cows at the Fence

Next we'll see if our 3D reconstruction techniques can tell us how tall the cows are.  If they look like cows, are as warm as cows, and are as tall as cows, we can be pretty sure they are cows, and not cow-shaped rocks or bushes.

Who does what?

Good job last week, folks.  We saw good teamwork on the "Build an animal shelter from newspaper" assignment that Helen gave you.  We also  completed another mission, the Animal Conservation mission worth 20 points, so we've moved it to the "Completed" column of our Kanban chart.

We also made progress on our model African Serengeti.  Tara and Macy got Google satellite images of the Serengeti and fitted them to our vibrating football table.  Nick, Alyssa and Lais then took several pictures from above the table, pretending the camera was a plane flying over the ground below.  They then used Autostitch, a program we used last year to fit microscope pictures together, to create a single bigger picture of the whole table:

With that job done we took pictures of Nick and stitched them together into a single bigger picture that I'm calling Nickenstein (this week's word).

The ability to put together small pictures into one big one will help us take closeup picture of animals and put them together into a picture of the whole area they cover.

Also last week we put together a list of jobs that needed to be done for our project and they included:
a) Fly the drone
b) Take pictures
c) Fit the pictures together
d) Train the computer to recognize the pictures
e) Count the animals in the pictures
f) Get pictures of model animals

You each voted for the top two things you might want to do and based on those votes we matched people to jobs.  One thing we discovered since last week is that if you want to fly a drone for anything but the fun of it (and doing research in a school doesn't count as fun even if it is) then you have to be 18 years old.  What that means is that flying the drone over a farm to take pictures of animals for our project can only be done by someone who's 18 or older.  On the other hand, you can all fly the drone for fun, so that's what we're going to do.  Let's hope for good weather at the end of the week!

Back to the jobs.  With flying the drone for research not one of the choices any more, we divided things up as follows:
a) Fly the drone for fun: Everybody
b) Take pictures: Neshaya and Lais
c) Fit the pictures together: Lily and Lais
d) Train computer: Lily, Alyssa and Olivia
e) Count animals: Olivia and Hannah
f) Get pictures of model animals: Nick, Tara and Macy

Also, Alyssa volunteered to do another important task

g) Investigate regulations around flying drones: Alyssa

As we mentioned last week, one of the researchers Olivia emailed answered her back and was super helpful.  He said:

Dear Olivia and the comet warriors,

Thanks for your email - it's great that you're trying to solve this problem. I put a folder in dropbox with some images from the wildebeest survey. The link to the folder is here (let me know if this doesn't work) - 

https://www.dropbox.com/sh/0r9dp6kmp667mmd/AAB0rBmKP2Tvz7HqpjHXZQbra?dl=0

There's also a blog article about how they perform the survey here - 

https://fzs.org/en/projects/serengeti-conservation/news-serengeti/count-wildebeest-serengeti-ecosystem/

The person who wrote that blog is Bryna Griffin, her email is bryna.griffin@fzs.org, and she might be able to provide you with more images of different animals and will probably be really interested in your research.

I should warn you I think what you're trying to do will be difficult. If you look at Fig. 2B in my paper you'll see that even people struggle to accurately count the wildebeest in these images! In my opinion the best way to improve the process is to focus on getting better information about the shape of the animal. You could do this by using a stereo camera or another technique known as structure from motion (https://en.wikipedia.org/wiki/Structure_from_motion). 

We've been trying to do this with a drone and testing it with cows too - I put some images from our tests in the dropbox folder. There's two images of cows taken at slightly different angles (im1.png and im2.png) and the height.png image shows the height of the cows by calculating the difference in movement of the pixels. 

I hope this helps with your project but feel free to ask me any more questions

Good luck

Please take a look at Bryna's blog post on counting wildebeest (https://fzs.org/en/projects/serengeti-conservation/news-serengeti/count-wildebeest-serengeti-ecosystem/) and see what questions we might have for her.

Fundraising and Mission Progress

Congratulations Comet Warriors on a successful yard sale on Saturday!  We raised $230 and got rid of a lot of junk at the same time.  It involved a lot of work sorting, selling, serving, and then cleaning up, so thanks to all those who participated.

We also made some good progress on our robot missions.  Olivia and Lily built an attachment to solve five different missions  - namely transporting the dog trainer to the research area, moving the veterinarian to the research area, doing the animal exchange, retrieving the seal-mounted camera, and putting the gecko on the wall.  Nick, Lais, and Alyssa then programmed it to do the Biomimicry mission,

Last week we made good progress on our seeing-eye dog mission.  Olivia built a robot cart to help the robot get over the gate.  We then tried to program it to complete the mission and got it to work once.  Now we just need to make it reliable!

On the project Olivia and Lily worked to get pictures of our toy animals for training vision systems.  Below is a shot of Daisy (the word for the week) from above.

 Tara and Macy worked with the electronic football table to develop a simulation of animals wandering around a field.

Our hope is that if we can train the vision system what toy animals look like, and have it identify them wandering on a toy football field, then we can work up to having it identify real animals in real fields.  To make that work we need to find some farm animals we can take pictures of, preferably by flying our drone over their field.  Does anyone know anyone who has cows or horses?  Hopefully lots of them?

Project Selection and Mission Progress

Last week we selected a topic for our project.  We had narrowed the choices to Wildlife Counting, Bee Colony Collapse, and Spectacled Bears.  It was great to see that numerous people had done some detailed research on the topics.  After some excellent presentations we voted to see which project we would do.  The vote couldn't have been closer, with Wildlife Counting coming out ahead of Bee Colony Collapse by 5 to 4.

In researching this topic, Olivia emailed the President of the Wildlife Conservation Network and asked him a few questions:

Dear Mr. Knowles,

    My name is Olivia O'Driscoll and I'm emailing you on behalf of my Lego Robotics Team.  This year we (and 70,000 other teams) are working on finding ways to improve human-animal interactions.  We spent a lot of time looking at the WCN website and talking to people in conservation and think we have an idea.  We're hoping that you could give us your thoughts on whether it would be worthwhile.

    A basic problem in conservation seems to be figuring out which species need help and whether the help they get is working.  Answering these questions requires knowing how many of a species exist and whether the number is growing or shrinking.  One paper we saw said "Accurate and on-demand animal population counts are the holy grail for wildlife conservation organizations throughout the world because they enable fast and responsive adaptive management policies." (http://journals.plos.org/plosone/article?id=10.1371/journal.pone.0156342)

    The people who wrote that paper came up with a way to count wildebeest in aerial photos.  They say it works pretty well, but still needs some improvement.  We think we might be able improve on it. 

    Since those people are scientists instead of conservationists we thought we should check with you before getting too excited.  We're hoping you could answer a couple of questions for us:

1) Do you think this is a problem worth working on?

2) Are there solutions that are already good enough and cheap enough?

3) Are there obstacles to getting solutions adopted even if they work?

    We know you're super busy but if you could give us any insights on this problem, or any others you think would be working on we'd really appreciate it.  Thanks!

-- Olivia and the Comet Warriors

He kindly got back to us and anwwered:

Dear Olivia and the Comet Warriors,

Thank you for your email and thanks for your commitment to wildlife conservation.  I apologize for the slow response as I've been traveling.

Before answering your questions I wanted to qualify that I'm not a conservation biologist nor have I done any of this work in the field.  I do support this type of work and my 22 years working in this space does give me a broad perspective.

So, to answer your questions:

1) Do you think this is a problem worth working on?

Yes.  In order to affect any type of change one must be able to measure the system before you take action and after the action is completed.  Counting is expensive, time consuming, and inaccurate. For example, Paul Allen spent $6 million counting elephants and the results took three years to come out, however, the results of those data are defining conservation actions across Africa for elephants.

A less expensive, more accurate way of counting animals would be of great value.

2) Are there solutions that are already good enough and cheap enough?

There are a host of solutions and you can probably find a number online.  They include, for example:

a) Aerial counting from planes

b) Using camera traps to see what walks by

c) Collecting scat (poop) and doing DNA analysis to figure our how many unique animals it represents and then using statistics to estimate populations.

d) Putting up "hair combs" on trees and then putting scent on them so cats (Canadian lynx, mountain lions, etc) rub them and then analyzing the hairs.

These solutions range widely in their cost and the value of the data.  Sometimes you only want to know if the animal exists at all.  Sometimes you want to know if there are "some, many, or a lot" of the species.  Sometimes, like with the Amur leopard, of which there are only 32, you want to know EXACTLY how many there are.

Some of the methods are good enough and cheap enough.  Many are not.

3) Are there obstacles to getting solutions adopted even if they work?

Yes, there are many obstacles.  Scientists can be set in their ways and have strong egos about how things are done.  Cost is always a consideration.  Using the same method over a long period of time is important so you can compare to previous results.  However, if there is a better and less expensive way to do this it would be eventually adopted and used widely for the great benefit of wildlife conservation.

I hope this is helpful.  I'm excited to learn more about your idea and if we can make it happen.

All the best,

Charlie

This should give us a starting point on our research, and an idea of what kinds of animals we might count and how accurate we need to be.  We'll talk more about this this week, and the word for the week is "Charlie".

An example problem would be to figure out how many zebras are in this picture:

We also worked on our robot missions.  For several missions we tried this approach:

1) See if we can do the mission with our hand.

2) See if we can design a robot attachment to do what we did with our hand.

3) See if we can attach the attachment to the robot and do the mission by pushing the robot.

4) (For this week) see if we can program the robot to move the way we pushed it.

See you all on Thursday!