czech english

SubT Finals

Robotika Team

After three years of development, DARPA Subterranean (SubT) Challenge teams will get the chance to compete in the Final Event being held at the Louisville Mega Cavern in Louisville, Kentucky on September 21-23, 2021. All teams have to re-qualify for both System and Virtual competition. Is Robotika Team going to be one of the 15 System teams and 10 Virtual teams? Update: 24/9/2021 — And the winner is ...


12th February 2021 — Final Event rules and schedule

DARPA announced on Wednesday 10th the rules for the final event, see Final Event Competition Rules (pdf). All teams have to qualify and the requirements are stronger, so see also new Qualification Guilde (pdf). Note, that there is not much time left — The Systems Competition qualification deadline is April 21, 2021 and the Virtual Competition qualification deadline is June 29, 2021.
So what is new? The robots have to be able to recognize 10 types of artifacts: 3 common (Survivor, Backpack, Phone), from tunnel (Fire Extinguisher, Drill), from urban (Gas, Vent) and from cave (Rope, Helmet). Moreover there will be 10th, so far unknown, artifact type. In order to qualify at least one robot has to demonstrate autonomous detection of all types of artifacts!
For the System there will be two Preliminary Rounds, each 30 minutes long. There will be 20 artifacts but teams will have only 25 reports. Only six best teams will compete in the finals. There the limit is going to be 60 minutes and number of artifact 40 with 45 allowed report attempts.
The Final Event of the Virtual Competition will consist of a Preliminary Round and a Prize Round each with a separate solution submission window. The submissions to the Preliminary Round will be evaluated and the results are expected to be announced the week of August 9-13, 2021. The top 10 teams from the Preliminary Round will be invited to submit solutions to the Prize Round to compete for monetary prizes. The results and winners of the Prize Round will be announced at the Final Event being held September 21-23, 2021.

4th March 2021 — Virtual qualification world released

Last night was released the Finals Qualification World for virtual SubT. In Czech I would call it „kočkopes” as it combines tunnel, urban and cave into one world. Simply from the very first crossing you can go to Urban to the left, or to Cave to the right, or straight to Tunnel:
The qualification task is easy — just correctly report at least one artifact. But is is hard, or it at least seems to be. The robots did not move at all in the first two attempts due to a small bug (#807, fixed). But the score is still 00 …
I do not mind leading the empty table, but 1 point would be nicer.
The robots have seen backpack in urban area, but it is „slightly different” to what was robot used to see in Cave, so they did not recognize it tonight. So hopefully tomorrow.
Two Freyja robots were capable to follow the world boundary in 15 minutes (simulated, it took almost 3 hours on CloudSim) but they both met under the hill and start talking how it would be great to have some snow and ski … no that was another story.
p.s. after another fix on OSRF side (#809, it looks like the base was not covered by signal) Robotika scored the first point and is the first qualified team to Virtual SubT Finals!

18th March 2021 — 10th artifact type?

Well, it is still not official but thanks to leak in Virtual SubT - PR826 it looks like that the mysterious 10th artifact is going to be some kind of treasure
I guess that by Friday this will be official but I am really curious how they want to force the System teams to detect it — well, they can force it but how to handle it? So I am looking forward for official info!

25th March 2021 — Finals Practice Worlds

OK, The Cube was really officially announced on Friday but I somehow overlooked it (as usual). So now there is a new CloudSim release not only with the 10th artifact but also the leaderboard contains Finals Practice 1, 2, 3 worlds.
Yes, you can see team robotika now, sometimes leading with zeros. But on the other hand the worlds were a little bit disappointing — they reflect old circuits: tunnel, urban and cave.
There is also new feature — fog! And that is going to be fun:
Fog in tunnel
Fog in tunnel
If you are curious what LIDAR sees in the fog then the answer is simple — nothing. Both Freyja as X4 drone turned around and returned.

3rd April — Virtual Fog

I must admit that the new feature fog emitter is a lot of fun. Camera does not see anything except white image, lidar readings are random and there are no sonars available on the virtual robots. A little bit promising are data from depth camera:
so now I am running simulation with suggested filter. We will see if robot Freya will navigate through fog dense so much you could cut it (Rakosníček) and discover drill artefact and score at lease one point in Finals Practice World 1. It is nice that this trap is very close to the start (follow the right wall for 1 minute), and it probably worked … unfortunately you cannot see fog in web-view. It just scored the point, so no longer leading with zero points in Finals Practice 1!
Here is the video of Freyja in fog in the Finals Qualification world … or, if you want, you can watch it from the beginning. How many artifacts you have you discovered? Backpack in Urban? What about the helmet in Cave?? No? Well, check the fog again. Yeah, you can see meeting of two robots which caused switch to alternative route … and a little bit of looping. The exploration was limited to 15 minutes (A900L) and then the robot returned to base.

6th May 2021 — Eight teams have qualified for Systems Final Event

… and Robotika is among them!
Yes, we were also surprised that only eight teams (instead of fifteen) made it to the Finals. On the other hand it is the elite (yeah, except us … but you never know. This is surely related to the pandemic + if everything can be canceled and the chance to win $2M is low …
So who are the competitors:
Note, that the registration to Virtual competition is still open, but you can already guess from the leaderboard who is qualified:
One point is enough to qualify … if you check both lists you will see that this time at least 4 teams will compete in both categories!

16th June 2021 — Kloubák and Skiddy in Cave Výpustek

Last weekend we had our first Robotika STIX (SubT/System Test Integration Exercise) for SubT System Finals. And it was a very nice change after almost a year (in some sense even more = since we returned from Urban Circuits) of virtual live only.
The meeting place was cave Výpustek in Moravia Karst with other famous caves like Macocha and Býčí Skála nearby. Cave Výpustek was formerly used as a hidden factory during war and later as a nuclear shelter. The main space is relatively easy for our robots because the original natural cave was modified to serve humans: the floor has hard flat concrete, lights are built in the walls and are used for tourists excursions, and the ceiling is high enough so you can walk around without a helmet. The most difficult elements were small chain guardrails saying "do not enter", which was not correctly interpreted by robots.
The software integration was difficult as expected. Since the last year, when System Cave Circuit was canceled, the focus was Virtual competition only and some hardware dependent pieces were no longer supported (or at least definitely not retested on real hardware). In time the basic operating system shifted forward to newer versions and the device drivers (RealSense tracking and depth cameras) too. Yes and it also took a while to remember all the "minor details" necessary to have everything smoothly running. But at the end both robots Kloubak K2 and Skiddy moved and autonomously navigated and explored the cave — the primary goal was reached. Note, that we were still rather recovering the old strategy from last year Urban Circuit than the latest algorithms from Virtual world (mainly terrain traversability and octomap mapping and exploration). This is going to be our next step.
Skiddy — this is our newest robot built by Pavel Skotak. It is a relatively small (0.4m x 0.56m x 0.35m) robot capable of entering narrow spaces. It is not the lightest (18kg) because it is based on four hoverboard motors (BLDC motors built in the wheels) which each weighs ~3kg. The robot was still "too fresh", not all sensors were mounted (so far front RGBD and tracking camera only, the rear depth camera was not integrated yet). Never-the-less robot was driving around, demonstrating its driving capabilities and also the weak points (connectors). The STIX was very important for the whole team to align and see the complete status first hand.
p.s. There were a couple visitors including MartinL — and now there is a slight chance that Skiddy won't be the last robot in the list for Platform Qualification which ends on 21st June 2021 (5days).

24th June 2021 — Czech STIX Býčí Skála

If you thing that subterranean robotics is only about robots, broken wheels, fried motor controllers and loose connectors then you are wrong. It is also about logistics, team coordination, catering etc. The last night was a perfect example. We were at the university testing communication between Jetson Nano modules with the main control computer, finished at 9pm, bought some groceries before the shop closed at 10pm and then drove "highway" D1 to Brno and cave Býčí skála. We arrived at 1am the next day and surely several people from CTU-CRAS-NORLAB team were still discussing ROS topics by the fire.
SubT is also about exciting (not necessarily positive) experiences. That come about 30 minutes after we arrived and evaluated options where to sleep. The best option (I would vote for it again!) was large open cave on tourist hike, about 300 meters away. Shortly after we entered (and probably woke up several competitors) the storm started to be much closer. It was fun, we could see the walls from "the large windows" with each lightning … but then very heavy rain started, which looked also OK, until I realized that I sleep in a stream bed. It was the beginning so still OK, moved to higher dry place. But then, about 15 minutes later opened waterfall directly on my sleeping bag. "Fun." These are experiences you will never get sitting behind computer at home and playing SubT Virtual track. (I can easily imagine Bilbo saying "we do not want any adventure here").
The cave is very nice - large, spacious, a great playground for all kinds of robots. Outside was temperature around 28C degrees and inside 8C degrees, so the very first „surprise” we have seen was LIDAR blindness, again:
And here are the views from front and rear cameras:

29th June 2021 — Team Robotika qualified for the Virtual Competition

We just received mail:
Team Robotika, Congratulations! After careful review of your submission, we are pleased to inform you that you are now qualified to compete in the Final Event of the SubT Challenge Virtual Competition.
So now the question is if all 12 teams now listed in Finals Qualification World will be accepted.
SubT Virtual competition deadlines
SubT Virtual competition deadlines

2nd August 2021 — Finals Preliminary Round

It is over, at least the submission. There were many changes you could still track GitHub including the FPR release (ver115) and team of 8 robots trying to score and push Robotika in the Top 10.
The evaluation is running now and according to SubT Wiki the next deadline is
  • 26 AUG - Prize Round Submission Deadline (there is no info when preliminary round should be completed … but I have seen it somewhere)
I see that I even did not publish link to the official announcement of 12 qualified teams. And there it is: Results from these initial runs will be announced by August 13.

12th August 2021 — Advancing to the Prize Round

Good morning,
The Subterranean (SubT) Challenge Team would like to congratulate you on placing in the top 10 of the Preliminary Round and advancing to the Prize Round of the Virtual Competition in the SubT Challenge Final Event!
With your team advancing, we would like to extend a friendly reminder that Prize Round submissions for the Final Event are due August 26th, 2021.
Please see a further reminder of dates of importance to keep in mind for the SubT Challenge Final Event:
  • Prize Round submissions are due August 26th
  • SubT Challenge Final Event will be broadcasted September 21st-24th on SubTv
  • The Awards Ceremony will be held on September 24th, where the winners of both the Systems and Virtual Competition will be announced
We will be following up in the coming days with an email providing further logistics regarding in-person attendance at the Final Event for you and your team. We look forward to seeing your Prize Round submissions!
Sincerely, The SubT Challenge Team

p.s. the official results are published here https://www.darpa.mil/news-events/2021-08-12

27th August 2021 — 2C or not 2C?

That is a question! Shakespeare would probably use different letter like 2B or not 2B?
You can tell that my mind is kind of broken after recent (and still running) SubT Virtual coding/testing marathon. Today is the Virtual D-Day, i.e. in 5 hours (not sure if I will finish this text any soon) 26 AUG - Prize Round Submission Deadline. It is AoE (Anywhere on Earth) which means 2pm CET time. And that should be noted even this blog is so silent this year …
Stress test, nice, I like that comes to my mind quite often, recently. It is phrase Camryn used in Cave Circuit (actually around 1year ago). If you forgot who she is then here is refreshing picture from the video:
So what happened last month(s)? I would start with nice pictures — feature I was against (it is optional to provide map data to the DARPA server) but it did not stop Jirka.
Even funnier was the „patching” when the browser did not display some tiles:
Timeout - switch to (regular) work and do the final release (SubT) …
p.s. C stands for center used in robot code names and "2C" meas prefer center for 2 seconds … and now the question is if to help Freyja robots to leave the starting area with this extra push

Game is Over

… so just to add (for myself) some reference pointers:
world:   Final Prize Round
image:   138467776890.dkr.ecr.us-east-1.amazonaws.com/subt/robotika:ver133
simname: ver133frr1
robots: 
  D390W2C1500LH3600W: ROBOTIKA_FREYJA_SENSOR_CONFIG_2
  B90W2C2400LH3600W: ROBOTIKA_FREYJA_SENSOR_CONFIG_2
  A0W2C2400RH3600W: ROBOTIKA_FREYJA_SENSOR_CONFIG_2
  C300W2C1500RH3600W: ROBOTIKA_FREYJA_SENSOR_CONFIG_2
  E600W900LH3600W: CORO_PAM_SENSOR_CONFIG_1
  H1980W600RH3600W: CORO_PAM_SENSOR_CONFIG_1
  G1920W600LH3600W: CORO_PAM_SENSOR_CONFIG_1
  F660W900RH3600W: CORO_PAM_SENSOR_CONFIG_1
       T4000: TEAMBASE

17th September 2021 — Robotika in USA (-4days)

Believe it or not, Robotika team is now in Louisville, KY, United States.
It was not straightforward as all travel from EU to USA is prohibited. You have to ask for an exception (NIE - National Interest Exception), which you may or may not get. We got DARPA invitation letter, which surely helped. Nevertheless confirmation that we could go we received on Aug 18th less then a month ago (We are pleased to notify you that a national interest waiver to Presidential Proclamation 10143 has been approved for your travel to the United States. ... Please note that an approval of a national interest exception does not guarantee entry into the United States. Yeap. Plane tickets, accomodation, car rental, robots, batteries, … no time for relax.
Who will fly over the Atlantic? Which robot we should take? What SW we should run on it? These questions were the most pressing while we were still finishing Virtual Finals. The member list self-reduced to 5 people from Czech Republic (Jakub, Standa, TomasP, MartinS and me) and Frantisek flying over from Maryland.
Most of the robots we left at home. Promising are Kloubak K2 and new Skiddy. We dropped even Maria living in U.S. as Frantisek is going to NASA Space Robotics Challenge awards next week (the day after DARPA SubT) and there is nobody who could take care of her.

18th September 2021 — Mega Cavern (-3days)

Well it was yesterday, we have seen it for the first time … and it is huge! (as expected)

19th September 2021 — Mega Trouble (-2days)

I am not really sure if it is a good idea to write down any report today, but … if not now when etc.
We are having troubles with rear motors and motor driver on robot Skiddy. As we have only 2 robots it is almost 50% of our robo-team, sigh. We spent non-trivial time because of
fw_version_major: 0 (uint8)
fw_version_minor: 5 (uint8)
fw_version_revision: 1 (uint8)
fw_version_unreleased: 1 (uint8)
vs.
fw_version_major: 0 (uint8)
fw_version_minor: 5 (uint8)
fw_version_revision: 1 (uint8)
fw_version_unreleased: 0 (uint8)
Thanks to Beda from CTU CRAS NORLAB team we have new 56V driver (replacement for fried(?) 24V version), but are the firmware versions above the same? What unreleased means? And what exactly is this version? Currious? It was 0.5.2!!! The configuration and API changed between these two versions which means more trouble.
OTOH there were also some good news — we passed E-stop check qualification and also the network test on staging area worked fine. We have also see the entrance for the robots. Deadly — steep curved turn left, so it will be the very first tests if robots are properly braking! BTW the mapping server worked too. The only detail was that when we were testing it with docker mapping server we could not see anything, because there was only one testing point. Even the corridor map from logfile was recognized by DARPA people, but we have not seen it …
What else? We defined internal "rule" how to select the one unlucky(?) person, who will stay in the garage. There are only 1+4 people alloweed in staging area and we are 6.

21th September 2021 — Robotika scores first point

Today (or better to say yesterday, it is 3am here now) we went through the first preliminary round. It was hard from the very beginning (as expected) as behind the gate is already steep (about 20 degrees) downhill slope terminated with sharp turn left. That is all you can see from the staging area. Surely enough also the WiFi signal even for strong directed antenae ends on the crossing just behind the turn. Unfortunately it holds also for LoRa (long range radio) device we use.
We are not allowed to publish any pictures from the internal course yet. The official recapitulation of the day will be broadcasted today (22nd September 2021) on DARPA SubTv:
(so I will put pictures here say in 24 hours once this official part is over … just for you to have an idea what to expect underground)
Kloubak K2 definitely did not want to go downhill (six times?). The good news was that although K2 did not like the slope (maybe also the transformations in the map?) it always returned back to the staging area so that we could try again. The robot may easily block the whole passage and no other robot (in our case only one — Skiddy) could pass it. After tweaking some parameters we "push" it down, it managed to navigate narrow spaces and picked urban entrance … it was funny to see crossing similar to Virtual Qualification … yes this sentence is maybe already leak, but it will be surely revealed soon and all teams already have seen it.
Skiddy remained at home — we wanted at least one robot to survive. More over Skiddy has broken rear right wheel:
In [43]: dump_errors(odrv0)
system: no error
axis0
  axis: no error
  motor: Error(s):
    MOTOR_ERROR_UNKNOWN_TORQUE
    MOTOR_ERROR_UNKNOWN_VOLTAGE_COMMAND
  sensorless_estimator: no error
  encoder: Error(s):
    ENCODER_ERROR_ILLEGAL_HALL_STATE
  controller: no error
axis1
  axis: no error
  motor: no error
  sensorless_estimator: no error
  encoder: no error
  controller: no error
… we suppose HW issues on wheel/cables/hall sensors … and hopefully it will be fixed today (the components were delayed by one day :(.
And how far K2 managed to go? It died in 9 minutes of 16 remaining (the preliminary rounds are only 30 minutes long, or short, and most of the time we spend convincing our robot to really go there):
0:07:57.612168 16 [2, b'\t', 0]
0:07:57.612415 16 [1, b'\x00', 0]
0:07:57.612754 16 [146, b'\x00\x00\x00\xc1\x001\xff\xbe', 0]
0:07:57.612943 16 [145, b'\x00\x00\x01C\x00J\x00\x84', 0]
0:07:57.613157 16 [148, b'\x00\x00\x00m\x00\x0f\x00\x1b', 0]
0:07:57.623772 18 b'\x10h&\xc1!3%\xd9\x1f\xf9\x12h\x00\x00\x01=\x00\x04\x00?'
0:07:57.624101 16 [131, b'&\xc1!3%\xd9\x1f\xf9', 0]
0:07:57.628947 16 [147, b'\x00\x00\x01=\x00\x04\x00?', 0]
0:07:57.639438 18 b'\x10\x00\x00-\x80\x10D\x00\x00\x89{\x10\x04\x00\x00\x03\x11'
0:07:57.639733 16 [129, b'\x00\x00-\x80', 0]
0:07:57.640092 16 [130, b'\x00\x00\x89{', 0]
0:07:57.640376 16 [128, b'\x00\x00\x03\x11', 0]  ... chybi 131, 147 = 2x 8bytes ... tj. chybi cca 18 bytes
0:07:57.652187 18 b'\x00 \x00A\n \x00!\x00\x12H \x00\x00\x00\xa0\x00\x18\x00:\x12\x88\x00\x00\x00k\x00(\xff\x8f'
0:07:57.652359 16 [0, b'', 0]
0:07:57.652590 16 [520, b'\x00!\x00\x12H\x00\x00\x00\xa0\x00', 0]
0:07:57.652765 16 [192, b'', 0]
0:07:57.652934 16 [464, b'', 0]
0:07:57.653134 16 [1088, b'', 0]
0:07:57.653329 16 [0, b'', 0]
0:07:57.653496 16 [856, b'', 0]
0:07:57.653685 16 [327, b'', 0]
0:07:57.668018 18 b'\x10h&\xc3!3%\xdb\x1f\xf9\x12h\x00\x00\x01T\x00\x0e\x00:'
0:07:57.668202 16 [1144, b'', 0]
0:07:57.668572 16 [833, b'\xc3!3%\xdb\x1f', 0]
0:07:57.668748 16 [1992, b'', 0]
0:07:57.668990 16 [832, b'', 0]
0:07:57.669227 16 [0, b'T', 0]
0:07:57.690521 18 b'\x10\x00\x00-\x90\x10D\x00\x00\x89\x8b\x10\x04\x00\x00\x03\n\x00 
                                                 A\x0b \x00!\x00 \x12H \x00\x00\x00\xe7\x00\x08\x002'
0:07:57.690757 16 [0, b'\x00:\x10$\x00\x00-\x90\x10D\x00\x00\x89\x8b', 0]
0:07:57.691058 16 [128, b'\x00\x00\x03\n', 0]
0:07:57.691859 16 [2, b'\x0b', 0]
0:07:57.692100 16 [1, b'\x00', 0]
0:07:57.692335 16 [146, b'\x00\x00\x00\xe7\x00\x08\x002', 0]
… Jirka already pointed out yesterday, that I should not share these cryptic messages even within the team but … maybe somebody will like the puzzle too . In other words there are about 18bytes missing in the stream which then causes wrong parsing of CAN messages, triggering parsing asserts and terminating.
As side effect we reviewed that messages 0x11, 0x12, 0x13 and 0x14 release the wheels and for command desired_speed = (0, 0) it actually releases the wheels! It is nice for transition of moving states but for flip() (K2 and Skiddy are symmetrical so you can virtually flip front and rear instead of turn 180 degrees) with 2 seconds(?) wait in steep downhill it is deadly.
Other detail is to fix reporting of Z-coordinate, which was not that important for us in Urban and Tunnel as the place was on one level (if we did not go up/down stairs or fly). Yeah, a lot of fun.
Today (22nd Sept) should be more or less copy of yesterday and tomorrow (23rd, Finals Prize Round) too as we will probably start from the last place (no big deal). Note, that results of other teams are not secret but they are also not official yet. One colleague went and ask most of the teams, so Explorer, CoSTAR and Cerberus all have 7 points, CRAS CTU NORLAB 5 points, Coordinated Robotics 2 points and Robotika 1 point. If you want to see pictures and some details about all competitors have a look at:

22nd September 2021 — Virtual Finals - partial results

If you watched DARPA SubTv today you already know the preliminary results for Virtual Finals. There were simulations for the Tunnel, Urban and Cave running each 3 times. Here are the drafts.
And here are the partial results:
and finally intermediate cumulative results:

23rd September 2021 — Game Over

It is over now. At least for Robotika. We started at early morning at 6:30am, E-stop check, move to the stage. There was basically no time to test the last HW/SW changes. Yesterday even encoders on K2 stopped working — the CAN module got lose probably as Kloubak tried to overcome some of the obstacles. Yes, we also fried couple of fuses. It started with 20A, then 30A and for the finals TomasP and MartinS changed it to 50A.
Skiddy got overnight new board filtering signals from the Hall sensors. It seemed to work OK in the morning (4am). These were all good news .
The run was similar to preliminary rounds — there was an artifact on the junction (Jakub would got the lottery prize, because he guess it in advance correctly). Nice surprise was view to Cave where was another artifact.
b'{"score":2,"remaining_reports":43,"current_team":"robotika","run_clock":727.3}'
… maybe we should stop here and go home. Actually Kloubak did and showed up in the staging area again.
For the next attempt we just re-run it, K2 properly picked left way to the Urban area, nicely navigated in the coridor to the left where we have never been before. Suprisingly it was available on LoRa all the time — not sure what changed for the finals.
You can see that K2 was moving back and fourth:
Now we are in the „jail” = Sequester area, that we would not be able to share with anybody the secrest we discovered in the cave. On the other hand information worth $2M … I understand them … we just do not have it .

24th September 2021 — And the winner is …

System Mapping results
System Mapping results
Note, that Robotika mapping was available only offline after the robot returned and the WiFi signal ended on the first junction.

Partners


Supporters