Robotika.cz » Competitions » SubT Challenge » Urban Circuit

SubT Urban Circuit

Author: Martin Dlouhý, 2019-10-11

Robotika Team

The DARPA Subterranean Challenge Urban Circuit is the 2nd stage of two years project. The expected event date is February 2020. Update: 3/3/2020 — The Artifacts

Content

11th October 2019 — Urban Circuit Team Qualification Waiver

Please help DARPA find a new underground research lair

Official announcement on community website

An updated Qualification Guide is coming soon, but we wanted to provide an update on the expected qualification deadlines for the upcoming Urban Circuit:

Systems Team Qualification: December 3, 2019
Virtual Team Qualification: January 3, 2020

New teams that are interested in participating in the SubT Challenge must first complete their Team Registration before submitting qualification materials: https://www.subtchallenge.com/register.aspx

Urban Circuit Team Qualification Waiver

(received by e-mail, On 10/10/2019 5:11 PM, SubTChallenge wrote:)

Tunnel Circuit Systems Teams,

We hope your fall is going well. This is a notice that your team has received a Team Qualification waiver for the Urban Circuit of the Systems Competition based on your successful participation at the Tunnel Circuit event.

However, please stay tuned for the upcoming release of an updated Qualification Guide. We do plan to release additional guidance related to team registration and requirements for qualifying new platforms.

We look forward to your continued participation in the SubT Challenge!

Sincerely, Elissa The SubT Challenge Team

9th November 2019 — Urban Circuit Environment Preview

13th November 2019 — Urban Circuit Site Announced

Info from community forum:

We are excited to announce that the DARPA Subterranean (SubT) Challenge Urban Circuit will take place Feb. 18-27, 2020, at Satsop Business Park west of Olympia, Washington.

…

DARPA is also excited to announce that new prizes of $500,000 for first place, $250,000 for second place, and $100,000 for third place will be awarded to top performers among self-funded teams.

…

In addition, we are announcing two new artifacts for the Urban Circuit. The new artifacts are gas and a vent. The gas artifact is a CO2-emitting device used to simulate a range of hazardous air quality conditions, such as a gas leak, poor ventilation, or fumes and smoke. Finding this artifact represents identifying areas that would be hazardous for personnel, including areas where breathing apparatus may be necessary. The vent artifact is a typical supply register commonly found in homes or work environments. Finding this artifact represents identifying potential areas with fresh air or an escape route to the surface.

SubT_Urban_Artifacts_Specification.pdf

Gas Artifact

Vent Artifact

p.s. if you do not know it, Washington state is on the west coast, far far away … map.

6th December 2019 — New Robotika Platforms

There are four new platforms developed by Robotika Team. They are planned for urban mission in Olympia in the middle of February 2020. What can we expect?

MOBOS

The already pre-qualified from Tunnel Circuit is small robot MOBOS:

MOBOS with RealSense

There is a new (huge) light — actually it is not that big, but the robot is relatively small/3D printed. Also beside RealSense RGB-D camera there is Intel Tracking Camera — we are really curious how well it will work.

Tank Maria

Maria is based on commercially available threaded platform. It is also small, comparable to MOBOS. It is planned to overcome small obstacles and enter narrow passages. It should be able to handle stairs (work in progress).

Tank Maria

Kloubák K3

The next step in Kloubák Division is K3 — 6 wheeled robots with two flexible joins. It is primarily designed to handle complex terrain. The platform moved for the first time 3 days ago (see video recorded in university corridors).

Kloubak K3 on stairs

Drone Hermína

Finally there is ongoing development of drone Hermína, which should be equipped with the same sensor as MOBOS and Maria: RGB-D and tracking camera.

Drone Hermína

The deadline for platform registration is January 17th, 2020. I hope that by that time we will post already some interesting videos.

7th January 2020 — Status 39 days to takeoff

It was probably not good idea to compete in both System and Virtual tracks. There is 23 days left to submit solution docker to Virtual Track, only 10 days for platform qualification and 39 days to takeoff from Prague. We are leaving on Saturday Feb 15, 2020 to Seattle and then we plan to drive to Olympia.

So what happened for the last month?! First of all we got a great present of 3 new laser range sensors from SICK, which are now mounted on new platform Kloubák K3. SICK, thanks a lot

There was a progress with Control Center development: LoRa (Long Range radio) is used for communication to send small status packets and commands. Today the robots also started to send localized and recognized artifacts via LoRa to control center … so there is a hope.

In meantime we also discussed Crisis backup plan and here are the points which came on my mind:

1) The robot has to recognize danger. If it does not it will die or hurt itself.

2) The robot has to be able to return home. If this fail we will not get any info what is inside and if it is safe to send more robots to the same area.

3) The robot has to be capable of some basic navigation skills: follow the wall, navigate into empty space, follow trajectory described by polyline.

4) The robot has to know its position in DARPA coordinate system. If it does not it is hard to correctly assign positions of, even manually, recognized artifacts.

5) The robot should be capable to recognize artifacts.

This is priority list where if all five points are satisfied then you are running fully automatic exploration. If not you can easily loose your robot. BTW there will be SIX SCORED RUNS with one Course Reset in the middle, so it is not good idea to break the robot or its sensors at the beginning.

Here is the schedule available only to registered participants:

Tuesday, February 18, 2020 0700 | Team Check-in 0800 | Team Orientation Team Move-in & Systems Check-out

Wednesday, February 19, 2020 Team Move-in & Systems Check-out

Thursday, February 20, 2020 Scored Runs Day 1

Friday, February 21, 2020 Scored Runs Day 2

Saturday, February 22, 2020 Scored Runs Day 3

Sunday, February 23, 2020 Course Reset

Monday, February 24, 2020 - Media Day Scored Runs Day 4

Tuesday, February 25, 2020 - Distinguished Visitor (DV) Day Scored Runs Day 5

Wednesday, February 26, 2020 Scored Runs Day 6

Thursday, February 27, 2020 Awards Ceremony & Technical Interchange Meeting

21st January 2020 — Urban Circuit Systems Team Call (25 days remaining)

We are running out of time. Fighting on both System and Virtual is quite demanding and deadlines are coming. The one for Platforms Qualification just passed 4 days ago. We submitted several videos of Kloubak K3, MOBOS and Maria with hope that it will be sufficient for acceptance. We decided to „park” drone Hermina — it was partially expected that this particular platform is too big step, but with combination of sick team member we should not spend more time on this. So now we are waiting for confirmation or rejection.

Last week, actually exactly a week ago, we had call with organizers regarding details for competition in Olympia (in less then one month). It is going to be interesting! When compared to Tunnel Circuit the start will be inside the building, probably couple floors above the ground (BTW I believe that there are also several floors underground).

Satsop is a huge Nuclear Power plant. There will be two courses (i.e. not 3 as I guessed sooner):

Alpha Course — L-shaped start with staircase on the right
Beta Course — large open space with entrance partially occluded by wall

Alpha Course

Beta Course

We can expect several downdrops like empty elevator shaft, stairs and platforms. The good news, I think, is that there will be mostly protective barriers mostly for people. Remember Safety First! and this is America … and there will be DARPA people collecting the robots and preparing the course …

Vertical Shafts

timeout — time to go to work …

3rd February 2020 — Robot Qualification (12 days remaining)

There is less then two weeks to the competition. DARPA accepted two new robotic platforms: Kloubák K3 and Maria. Eduro, Kloubák (K2) and MOBoS where already qualified to Tunnel Circuit. Drone Hermína was rejected: The videos were missing aspects of both the mobility and e-stop required tasks. However, you are welcome to qualify additional platforms for future events including the Cave Circuit and Final Event.

It was necessary to upload pictures of all robots so here you can see new pictures for at least 3 robots currently living at the Czech University of Life Science in Prague:

Eduro

Kloubák K2

Kloubák K3

Yesterday I also received the latest picture of robot MOBoS, which was (unintentionally) testing fall from the stairs — luckily everything survived except the 3D printed cover. So what PavelJ did? Just re-printed it … and surely add some advertisement …

MOBoS with new 3D printed cover

MOBoS Realsense cameras and light

14th February 2020 — Leaving tomorrow …

It is hard to accept that we are already leaving to Olympia/Washington USA tomorrow. There has been non-stop development last two weeks mainly porting SW changes from Virtual Urban Circuit to System Urban Circuit. The Local Planner is optimized now to be capable to run on low power computers (robot brain), return_home() takes into account robot symmetry and works for articulated robots like Kloubak K2 and K3, etc.

Here is at least our group photo of people testing last Tuesday at the University of Life Science in Prague. It is not the whole team and the robot Maria is also missing (Maria lives in USA).

Robotika team photo at CZU

15th February 2020 — Montesano and missing bags

OK, good news is that the whole Robotika Team for Urban Circuit is now at Montesano — small house about 20 minutes from the test area. Bad news is that Air France did not manage to transfer two or our large bags in Paris, so we are missing wheels for Kloubak. The good news is that it looks like Kloubak K2 could be completed. The bad news is that for some unknown reason I had to remove sealed lead acid batteries from robot Eduro and now, after reconnection it does not work. The time difference is 9 hours, i.e. 5am back at home … and it is probably not good idea to do anything potentially dangerous.

There were two towers on the way from Seatle to Montesano (approximately 2 hours of drive) … and that is the place were we will compete next week. Hopefully by that time we will get missing wheels and other parts stored in the missing bags …

Two Towers

Not working Eduro

Building K2

16th February 2020 — Eduro and K2 back in the game

Eduro power is hopefully fixed. It looks like that one connector on one battery was lose when the battery was inserted and mounted inside the robot. Also robot Kloubák K2 is assembled and it already passed the first calibration tests. There are some issues with joint angle calibration — it should be fixed tomorrow.

The weather was nice today — it rained couple of times but interlaced sunshine just leveraged the beauty of the surrounding forest. There is also river and several creeks nearby … nice relaxation to full time programming.

Nearby forest

p.s. here is video from packing Kloubak K2 and K3:

17th February 2020 — Presidents' Day

Today was Presidents' Day which we celebrated by testing robots outside … it was freezing in the morning but sunny later on.

Now I can list here views of all robots. Well not really as the bags with Kloubak K3 wheels will arrive hopefully tonight?!

Kloubák K2 view

Eduro view

MOBoS view

Maria view

Summary — all robots moved, MOBOS and Maria used hybrid system with ROS, Eduro collected data with slope lidar and K2 tested flip for return from narrow passages. Tomorrow we start at 8am with meeting in Olympia. By that time the robot K3 will be probably still in pieces …

p.s. we still do not have two missing bags with K3 wheels, extension cords, small monitors for Jetson Nano, and tools 60 hours after we landed …

18th February 2020 — Team Orientation, Move-in and Systems Checkout

Today officially started SubT Urban Circuit/System Track. We met with the organizers in the hotel in Olympia in the morning (8am, but no big deal because most of us were awake since 4am). Robotika had scheduled Move-in at 10:30, but we were still hoping for missing bags … and believe it or not they arrived!

By lunchtime Kloubák K3 was ready for action … good job, guys!

Cold morning

The Base

Forest

The weather was marvelous today, but instead of hiking we went to see the concrete structures in The Power Plant, what a crazy folk …

Outside

Inside

Robotika flag

All our five robots passed Systems Checkout so the next step is already the competition. The first run for Robotika is scheduled on Thursday 20/2/2020 at 10:30 (11-12).

19th February 2020 — Alpha Strategy

Tomorrow is our first competition run in Alpha Course so it is about a time to prepare some alpha strategy. What we have seen yesterday did not make us happy: the stairs on the start on the right side are quite steep. It looks like 45 degrees, but it should correspond to DARPA measures we got two weeks ago. There are two floors i.e. no go even for Kloubák K3. Moreover the hand rail starts at about 50cm from the ground so most of our robots could drive under it and fall down tens of meters.

So there is only one way and that is left. Current SW supports only one choice of direction at the beginning (yeah, I know that it can be easily(?) changed, but let's start with pieces we have now), so let say that all robots will follow the left wall. Five of them. Hmmm.

The strategy from Virtual System was to sacrifice two robots (left and right) for deep exploration with the risk that they will never return. But if they will, they will probably bring some „treasure”. One direction means one robot. My favourite would be Kloubák K2, which was already successfully fighting in Pittsburgh. Time I would set to 20 minutes of exploration, hopefully at most 20 minutes for return and remaining 20 minutes for analysis in the case of success.

The next robot I would send Eduro maybe with one minute delay. It would follow Kloubák K2 and record what is going on. It should return sooner, say 10 minutes of exploration = 20 minutes round trip. If it will observe K2 dead we will try to stop it remotely (not to harm itself even more). BTW when the robots are terminated they should turn into LoRa re-translation stations so that we at least know where is which robot. (TODO set of starting scripts specific for each robot)

Then we have MOBoS which which was able to explore some reasonable distance. It is small and it can enter small passages, but if it fails it can also easily block the return passage. In any case it is the first candidate for nearby exploration if both Eduro and K2 fail.

Then there is Maria — tracked robot. It is newbie and has once a while problems with loosing the tracks. At the moment it become road block (the passages towards the gate are not very wide, say 2.5 meters I would guess).

Finally Kloubák K3 which we should not use tomorrow for the stairs — that would be the end for 97%. It is large and it can easily block the road. Maybe we should use it to protect the staircase at the start area first? Without the climbing enabled it can easily block all robots … my first though is to send it at 0:40:00 if K2 returns.

The team is slowly getting up now, so we will see if they stone me to death.

Comments, including the rest of the team oversees, are welcome …

p.s. there is a new IEEE Spectrum article: How Robotics Teams Prepared for DARPA's SubT Challenge: Urban Circuit

20th February 2020 — Robotika scores the first point!

Our first competition run was live nightmare, as expected. The strategy described yesterday slightly changed — we used Kloubák K3 not only to build the protection wall for returning robots but also as holder of WiFi directional antenna. We bought 50 feet (approximately 15 meters) long Ethernet cable in local shop and plug it in the other WiFi router. The connection was good comparable to Pittsburgh/Tunnel Circuit.

We changed the order of robots:

the first started Kloubák K3 (osgar.go 4 meters) to build the wall and establish WiFi coverage
the second started Kloubák K2 along the left wall with planned exploration time 20 minutes

And then I saw it — just a perfect not day but a trap … well, I am not allowed to talk about it, I almost forgot. There are two more days when teams with higher rank from Pittsburgh will explore the same course, so later.

The third robot was MOBoS instead of Eduro as it should be able to overcome the surprise. It did not and shortly after the left turn it hit the wall.

So as the fourth robot we sent was Eduro anyway. It started on predefined L-shaped trajectory before switching to follow the right wall. It started nicely except the navigation trajectory was not long enough and the nearest wall was a column. Eduro started nice pole dancing. „Funny” enough the LoRa device did not work, so GoHome failed to be transmitted. Sigh. But Eduro was on WiFi so it could be semi-manually (follow trajectory with local planner) navigated back through the Stargate. In the second run it was caught in the trap, as expected.

The next was Maria started and went all the way to the next wall. It would hit it if I would not kill it in time.

So the very last robot was Kloubak K3, the hero of the day. We returned it back (osgar.go -5 meters), patched the SW (yes, it was not really ready for the task), disabled downdrops so that it would enter the course … and it went.

Later we could hear the noises like chainsaw and on camera I saw the sawdust!

That was the end of our last robot except beside the sawdust I noticed also the artifact. And we got the point! Hurray!

The leader board on Feb 20, 2020

Well, it was not great, but …

all robots survived and none fell down the hole
we scored and there is at least one team which did not
we know what to improve for the next iterration

p.s. we live in the woods and the internet is really bad here (but the nature is beautiful here) so I cannot easily create screen shots from the live DARPA video available on Subterranean Challenge Urban Circuit.

p.s.2 known bugs (now) — MOBoS and Maria failed because of --start-paused parameter which in combination of follow trajectory … did not work.

21st February 2020 — Beta Course - 1 point

The Game Level 2 is Over — well, it was not really next level, I should rather say that we (Robotika) were competing in Beta Course and scored 1 point. We should be happy about that but somehow our expectation this time was higher. Why? Well, all other teams scored at least one point in Beta Course too …

Leader board 21/2

source: https://twitter.com/DARPA/status/1231057181145010177

The situation from Pittsburgh repeated today again when I have seen the artifact but I was not able to locate it. (After home review we have not seen anything there so the answer could be that the artifact simply was not there so no wonder the answer was always False). It was detected by Eduro which returned to WiFi signal but the trajectory was „broken” while the first corridor looked straight. I tried to compensate this error from Kloubák K2, but as I said it was not artifact but only some reflection.

So what went fine and what went wrong? First of all Kloubák K2 was overnight upgraded to handle low obstacles from RealSense as the top priority. The priority kick was from Alpha Course, Jetson Nano running ROS was talking to OSGAR over ZeroMQ … it was the plan but the TODO list is quite long and every mistake will surely show up the next day. It worked, actually it worked very well! I had some doubts so I sent Kloubak K2 as the second that Eduro (old untouched code) will pass the narrow doors, but … Eduro did not like it and returned home. :-(

Kloubák K2 was sent and returned home too … hmm. So we went to smaller robots: MOBoS and Maria. MOBoS went nicely through the passage and start to explore small rooms on the right side of the corridor. And then it died.

Maria tried the same, but returned home again. Solution? Reduce the walldist to 0.6 meter (as it was on MOBoS) and re-try … and it worked. … but after Maria had seen the easy artifact turned to some small room left and died too. (*)

For the Eduro run we used the same „trick” as with L-shape start in Alpha Course — here the trajectory was set to 50 meters straight to push exploration in desired direction. The robot was exploring various wet areas with deadly machines (I wish I could place here the picture … maybe after the embargo I will revise this blog).

In meantime we sent Kloubák K2 to explore left side … it explored about 100m and dies on some small stone in a small room. (**)

DARPA video - Urban Circuit Round 1, Day 2

from yesterday:

SubTv - Urban Circuit Round 1 Day 1 - DARPA Subterranean Challenge

(*) just a couple minutes ago PavelJ discovered the reason. Both robots run on ROS and used names within the local network, but as soon as the connection was broken then names were no longer valid and the ROS nodes stop to communicate … and the robot died.

(**) Jakub just plotted Gas values and … yes, we have found another artifact, but did not report it:

CO2 graph from Kloubák K2

(no longer censored)

Guess where was it?

22nd February 2020 — Run 1 completed

Today we had „day off” (only the best teams from Pittsburgh competed today). The short rest was nice, some of us even slept, but unfortunately (?) we have to get back to work if we want to have a chance:

Leader board for Run 1

We have to end up on the 6th place in order to win some money from System Track. This means to score at least 2 points per run in order to have better score than Marble team from Colorado. And as we remember from Pittsburgh, the second run will be harder and all the teams will take higher risk (maximum for each Course is counted). Sigh.

p.s. we watched video from yesterday — teams Explorer and CoSTAR and at least the walking robots from CoSTAR were very impressive. Their robots walked down the two floors, and it turned out that they cooperate with Boston Dynamic which is „partner” of CoSTAR (NASA, JPL, …).

p.s.2 tonight we had Teams Mixup evening (?) and we learned that most of the teams teleoperate their robots. For example Cerberus have also walking robots, but when the base robot with long cable lost connection the robots were automatically directed to return home autonomously, but failed …

23rd February 2020 — Beta Strategy and Reset Day

Today is the Course Reset Day when DARPA will build and break the walls, reposition artifacts and prepare new traps for the robots. Robotika team starts from the almost last position (9th) so we will start first in Beta Course tomorrow. How do we change the strategy when compared to previous run?

Well, the original plan was to start long range exploration with robots Eduro and K2 — right and left side of the unknown space and ignore small entries on both sides of the corridor. Then Maria and MOBoS would explore that part, but … it looks like the motors on Maria are really fried (both of them) so missing line was fatal:

"virtual_bumper_sec": 10.0,

or better the one from K2 and K3

"virtual_bumper_sec": 10.0,
"virtual_bumper_radius": 0.5,

(the default value for virtual_bumper_radius is

virtual_bumper_radius = config.get('virtual_bumper_radius', 0.1)

… hmm, I expected meter … well, definitely explicit is better then implicit here)

What is Virtual Bumper — it goes long time back to Cleaning Contest in 2002 when our robot Berta won the first prize in Switzerland. There we change the disadvantage of weak motors into advantage: if the motors should move (PWM above some threshold) but encoders do not show any change, then you are probably stuck on some obstacle. Here was Maria stuck on some concrete block and even the full power to turn was not enough. :-(

Here I am surely the one to blame because I did not force enough the new ROS-hybrid robots to be merged into master. There would be higher chance to spot similar mistakes…

So back to strategy — if we loose Maria then MOBOS probably has to handle both sides of the corridor with narrow passages. There is also discussion if to use 4WD on K2, which will make odometry from encoders less useful, but RealSense tracking camera can compensate that. K3 drives 4 wheels by default and remaining 2 uses for passive odometry … and we do not plan to drive all 6 wheels next two days.

Well, we start at 9:30am and competition run is from 10:00am, so wish us luck tomorrow.

Robotika team with robots at the nuclear power plant

24th February 2020 — Run 3 - A Nightmare on Elma Street

What could possibly go wrong? And what could be worse? Well, I have witnessed it …

At the beginning everything went fine — my laptop on the right, Zbyněk's laptop on the left, live video stream from GoPro mounted on K2 via Honza's phone. Zbyněk prepared tab with automatic rsync (downloading logfiles from currently online robots) of all robots (with optional preference with one), the second tab contained Jakub's tool for analysis (heat map) constructed on all collected logs for WiFi Phone artifact signal and CO2 level. Then I had 5 terminals for visual inspection of downloaded logfiles, one for each robot. There were also two spare terminals ready for „anything”.

One minute to the start I lost console to Eduro. Then K2 was deactivated … no way to even start the robots! Then I noticed WiFi signal from Phone so I tried to report artifact. That failed too!!! It is hard to score if you do not have connection to DARPA scoring server!

Then I switched to Zbynek's notebook, which was connected, started Eduro and synced logs. Eduro went all the way to L-turn and then I lost connection.

The second was K2 (right side this time) went to the column (on the right side), turned around and returned home! That was nice. WiFi did not work, the robot could not connect to the base station even as it was already in the Pit Crew area. The only solution was to connect via cable and downloaded logfile. Restart took 15minutes and the connection still did not work. Now come the true autonomy, K2 was started via cable (Jakub) and the only connection was LoRa (robot position and automatically detected artifact).

Surprisingly K3 was on WiFi and could be started — went fine through narrow doors and disappeared somewhere near Eduro. Yeah, I forgot MOBoS, which after restart (there was some problem with LoRa), turned to narrow passages, and that was it. It seems that the problem with ROS and lost connection repeated again — we tested it home but maybe the WiFi was not off long enough or … nobody knows. There was also a problem with tracking camera (as we have seen now at home), which was failing probably due to partially occluded view by new bumper.

Maria with replaced motors moved a little bit, starting from the last position. It moved, turned to the first hole and stopped (the motors are for lower voltage and weaker).

Summary: 1 point for artifact reported directly by the robot.

Leader board 24/2

Robotika at Beta Course

Kloubák K3

Czech University of Life Sciences Prague

SICK

K2 safe return home

25th February 2020 — Run 4 - The Last Strike

OK, it is over now — our final score is [1, 1, 1, 1]. The maximum is taken from both runs, i.e. [1, 1] … and because we reported artefacts typically late we are going to be the 9th position (the last place from the teams scoring 2 points).

No robots were hurt during this run (*)

(*) one exception is broken right front down-drop on K2

p.s. updated leader board:

Leader board 25/2

SubTv - Urban Circuit Round 2 Day 2 - DARPA Subterranean Challenge

The strategy was the same as first time on Alpha — move in Kloubák K3 to provide better signal and then start one robot a time exploring each side …

K3 as mobile base with antenna

K2 entering the search area

Kloubák K2 made it all the way to reactor … and enter it (no signal zone). Note the 1/4 coverage of the whole area. Yeah and the zero score 10 minutes to the end too. We scored 1 minute to the end …

K2 in reactor

Seach coverage

Kloubák K2 stopped due to CAN failure (unexpected message E4 00).

What is not visible on DARPA SubTv is failure of Eduro … it returned twice (?) and the problem was with RealSense Tracking Camera. SICK lidar was in the field of view of tracking camera. It was the best part to track when the light was not very good and only image of lidar was very stable … basically it created map all around Eduro without any chance escape.

26th February 2020 — Lessons Learned

Today we have day off while the best teams are still competing. And in the afternoon is going to be excursion of the Alpha and Beta courses. There will be final announcements, awards ceremony and technical interchange meeting tomorrow so it is time to summarize lessons learned we could share with other teams. Moreover I am tempting to let František speak … he did quite well, see DARPA TV after our last run.

OK, so what is the main lesson learned for our team? Or for myself? I think that we are on the right track and that is the combination of both System and Virtual Tracks. This implies full autonomy and definitely not great results in System Track. My team hates me for that and maybe it is not the right competition if teleoperation is allowed (or worse preferred) … „why not to switch to remote, pick some points and win at least some money?” they said. I do not know, maybe it is because I have seen it before :-(. It is not only my hate to simulators but also to joyride (robot controlled by joystick) … it is tempting and it takes huge effort to return back to what we really need to do and that is make the machines fully autonomous in situations when the human is not available. And that will happen. And that is the interesting part (for me) — then you can explore kilometers underground with relatively cheap equipment, no need for communication (yes, it is nice to have it but you have to be able to survive completely without it).

I have seen the effect yesterday afternoon after the last run, when the team wanted to do something — so they connected some ROS modules and extended OSGAR not to only send desired velocity commands but also to accept them. As a result they were able to run both Kloubák K2 and K3 on remote outside in the terrain. They all enjoy it only I was sad. It was like that they finally could do what I „prohibited” … but it was in the right time so let them have also some fun at the end. (I do not know how long I am able to hold my position though)

I was going to put some clip named Full (of foolish) autonomy which would be collection of bits from SubT Urban:

Eduro dancing around the pole
Eduro hitting the wooden barrier, repeatably
K2 going there and back, there and back, many times and the down-drops triggering alarms on the low barrier
K3 going going across the wooden barrier (at least the front wheels)
MOBoS trapped in a small room failing to find the way out
Maria in flames when was stuck and Virtual Bumper was not properly configured

All these scenarios could be easily resolved by operator with remote access. Except the last case (or maybe even there) we had connectivity and video input … so maybe it is really stupid to let the robots in the traps? But on the other hand it was lesson hard to forget and pushed us to complete integration with RGBD RealSense camera, which was already prepared on most of the machines but was not on the top of the priority list. Total failure in the fist run change that and that is how it could work.

I should list also some cases when autonomy worked nicely:

K2 entering reactor via Stargate
Smooth navigation through the doors on Beta Course, all robots
K2 finding phone artifact in a small room (without connectivity)
Eduro returning home on LoRa call, nice, unsupervised

What technical lessons we learned? Primarily it was testing of Intel RealSense depth and tracking cameras. They work very well, but …

USB3 noise harms WiFi communication (see USB 3.0* Radio Frequency Interference Impact on 2.4GHz Wireless Devices - White Paper)
tracking fails when parts of the robots are in field of view
water on the floor creates false obstacles/holes

In particular the first point was if we want to navigate safely or if we prefer connectivity to the robot. We preferred safer robot.

Another hard stressed sensor was IMU. We have seen it to fail before, but that was due to some checksum error partially caused maybe by nearby RealSense. But what we have seen in both Alpha and Beta courses during the 3rd and 4th run was change of validity flag from 1 to 0, on which OSGAR reacted not to send the IMU angles into the system. As a result the robot was orientation blind and that was fatal for the application. I still guess that it is rather recommendation flag not to trust the orientation, but … what can the robot do deep in the dungeons?!

I am glad that now we have two independent position system: IMU + odometry (as we have it up to now), and RealSense tracking camera. Data for both were recorded and we hope to fuse them into more reliable output. Imagine failure of the light when the robot has to return blind … that is situation when the tracking camera fails, for example.

Another success was intelligent sensor box. It consists of Jetson Nano, RealSense tracking and depth cameras. It is running ROS and connected to the robot via Ethernet cable. This whole unit can be placed on any robot (requires some configuration changes though) and provides filtered scans and map, which can be used by the application. The advantage is that it can be developed once and deployed on all robots by simple copy of the SD card (Jetson file system).

Also LoRa worked fine, mostly. Back at home we observed several LoRa failures in the building after few turns at the university corridors, but here in Satsop building it worked fine. (the reason could be that the robots re-forwarded the messages, which I did not check yet)

The control center improved too. I could see positions of all robots (yes, it looks like from old movie Alien but I can still track „where the beast is”). Also the unified coordinate system (DARPA frame) worked much better. There is still TODO, which is global time (should start from 0 at the beginning of the competition run). Moreover robots should report their health (running, paused, stuck, burning, …).

Jetson Nano has GPU on board and one of the promises was that it will handle detection of artifacts. It was high value for us as deep autonomous exploration could be or better to say, typically is fatal and terminates due the system failure. If the robot did not report anything (even hints) on its own, the team does not score.

I could also mention some software lessons. This time we tried hard to merge intermediate results in the common master branch. It was definitely not smooth, but I have much better feeling that both Jakub and PavelJ were fully involved and the pull-request and review process run even now when no time is left. Yes, that is also the time, when it is the most important and some stupid bug could endanger the whole team.

Note, that some contributors are not mentioned here as I (m3d) integrated their patches.

There is successfully running hybrid system (OSGAR + ROS) also in System Track (in Virtual was was must have). Newly also the control program is distributed on two computers APU and Jetson Nano on Kloubák K2, K3 and Eduro.

Alright, the list is long enough I think . It will not fit in 15 minutes presentation anyway.

27th February 2020 — Awards Ceremony

Today is the Awards Ceremony in hotel in Olympia (starting at 9am PCT, i.e. 6pm of CET). The winner of System Track is clear: CoSTAR — I highly recommend to watch the last video … they are good!

Yesterday we had an opportunity to see both the Alpha and Beta Courses … and then all of us dive into a deep depression (DiDD). The turbine building had approximately 100x100 meters, 3 floors, and it would be impossible for human to explore it even with perfectly working permanently online robot. Whoow, what a place! No wonder there were shot several movies.

Well, back to the competition … here is the final (still not confirmed) leaderboard:

Leaderboard 26/2 - FINAL

Note, the 3rd position! Yes, Czech Team CRAS-CTU managed to score 6 points in their last run and reached very nice position after CoSTAR/NASA/JPL/Boston Dynamic and Explorer/CMU. Congratulations! (Yes, $500 000 prize is surely also nice)

The official ceremony was live on DARPA SubTv.

SubT Urban Teams Photo

28th February 2020 — The Cherry On The Top

There was one more highlight besides learning details how our competitors (CMU, NASA, Boston Dynamic, …) solve the problems. Yes, the success in Virtual Track also counts.

Thanks to Ross Read we could visit the SubT Urban Challenge location one more time. This time we could see it also from the top! Yes, the view from inside the cooling tower was also incredible and the sound effects hard to forget.

I would like to thank Satsop for this great opportunity … and the climb was definitely the Cherry on the top.

The Tower

Two meters deep basin with gold fish

Blue sky from inside

The stairs

Olympia mountains

The reactor and turbine building

The Climb

Inside chimney

On the top

p.s. I would like to thank DARPA … it was incredible experience and I believe it will push Robotika team another step ahead.

3rd March 2020 — The Artifacts

The game is over so it is time to review what robots have seen in the four runs (hopefully we will also create and upload some videos).

Run 1 (Alpha)

I would start with front camera view of Kloubák K3:

Barriers, Kloubák K2, Eduro and Backpack

These are the deadly barriers invisible to 2D lidars. You can see Kloubák K2 on the left, Eduro in the middle and if you look carefully also red backpack artifact on the right.