Lack of luck in RoboCup 2011

Two weeks after coming back from RoboCup 2011, finally I could manage to write a bit about Rescue Robot League.
This year, the league had two major absents: Adam, the league chair who has been organizing the league since its beginning and the team Pelican United, the best organized and very well known Japanese team with an exceptional track in USAR robotics.
In comparison with Singapore, the teams mostly had a good progress in autonomy. Two teams (MRL from Iran and CASualty from Australia) brought more than one autonomous robot in their missions and German teams – which have always been frontier in robot autonomy – demonstrated their recent works in 3D mapping. One team (Hector Darmstadt) successfully migrated to ROS and several other teams had started to do. In contrast with autonomy, I didn’t see anything new in robot mobility or mobile manipulation – maybe due to the absence of the Japanese teams, Pelican United and SHINOBI.
My own team, AriAnA, had really bad chance in Turkey. Unfortunately our luggage got stuck in Turkish custom due to a dull mistake in Iranian custom – they had forgotten to include the packing list which we had provided. Finally we had to prepare a list of every single device in each box to get them from Turkish custom. Unfortunately we received the luggage within the preliminary round and we had to set up the robot while other teams were competing.
As the winner of Best in Class Autonomy award in Iran-Open 2011, we were very optimistic to get a good result in RoboCup as well. We’d built a new high mobility platform (DELTA) as a part of our recently funded project, AMIR (Advanced Mobility Intelligent Robot) to pass yellow and orange arenas. We had also developed a multi modal navigation algorithm after Iran-Open competitions to cover drawbacks of both wall-following and traditional frontier based exploration algorithms. Unfortunately we couldn’t evaluate our new approach in Turkey though the robot was perfect in action. Maybe we should try our chance somewhere else!
At the end, I was selected as a member of Organizing Committee of Rescue Robot League in RoboCup 2012 Mexico. This was the only good news we brought from Turkey.

Winning “Best in Class Autonomy” award of Iran-Open 2011

We’ve just arrived from Iran-Open 2011, the biggest regional competition of RoboCup in Iran (and probably in the world). Thanks to the organizers, the competition was perfect in terms of schedule, facilities, internet connection …
Like in previous, Iran-Open 2011 hosted several non-RoboCup leagues (e.g. Demining, Quadrotor, ROV) besides all the leagues of RoboCup both in senior and junior sections. In the Rescue Robot League we were waiting for some foreign teams (from Japan, Thailand, Malaysia, etc.) but, unfortunately only one of them (from Mexico) could take part in the competition.
Comparing to the last year, the famous Iranian teams (like MRL, YRA, Pars …) were not as good as I expected. Probably they were at middle of their developments or it was solely because of the effect of long Persian New Year holidays!
Our own team, AriAnA, participated with only one autonomous robot – since we have started a new research project called AMIR (Advanced Mobility Intelligent Robot) after reorganization of the team. Although we had some difficulties with the temperature sensors, the system proved its robustness where we didn’t observed any serious problem in hardware or software layers.
As our first experience with an autonomous robot, I think we were successful in Iran-Open 2011 in which we could win the “Best in Class Autonomy” award!
Thanks to all my existing and former teammates for their great work!

First Place of Rescue Robot League at AUTCup 2010

Last week we took part in AUTCup 2010, a national robotic event; but this time with some quests from countries like USA, Germany etc. Generally, the competitions consisted of two categories: RoboCup Leagues (all the simulation leagues with most robot leagues) and Non-RoboCup Leagues (e.g. Mars Rovers, Deminers etc.).

The Rescue Robot League was very similar to Iran-Open competitions. Although the team MRL was absent, all other famous Iranian teams made it a very challenging league. The arena was a smaller copy of RoboCup 2010 rescue robot arena with several steep roll ramps and more difficult red arena step field.

Since our managing team had decided to participate with minimums; we registered only 5 members (usually 3 of them were ready on-site) and utilized BETAII, one of our tele-operated robots which had several minor modifications after coming back from Singapore. We also had another change; our former operator, Mehdi (my older brother) was not able to attend at AUTCup and I had to control the robot myself. I was well familiar with controlling BETA using Xbox 360 joystick but, now we were using Logitech freedom and I needed to test this new controller. Thanks to my teammates, I had several test & practice runs in setup and preliminary days. Now I was ready to do almost everything with robot at final rounds.

As our competitors said, we had two dramatic final rounds in which we scored all the victims on elevated floor and inside the car. In the second final round we all knew that we won’t win the first place unless I open the door of a bucket using manipulator and identify its victim. We practiced and I found a trick to open the door using our 2DOF manipulator. Fortunately everything went fine at the final round and I could do what I practiced!

Thanks to all my existing and former teammates who made it to become true!

Man-packable rescue robots rather than Superman-packable!

Obviously rescue robots should be light enough to be carried by rescue personnel to the disaster zones. US Center for Robot Assisted Search And Rescue (CRASAR) emphasizes on the usage of man-packable and more specifically “shoe box size” robots in Urban Search And Rescue (USAR) missions.
However the existing rules of RoboCup Rescue Robot League (RC RRL) do not encourage participants to develop small size and light weight robots. If you take a close look at the participating teams of RC RLL, you will hardly find man-packable robots. In my opinion, this is only due to the lack of “reward/penalty” mechanism for such a crucial parameter.
Certainly the size and weight of a robot can only be decreased at a cost of decreasing its electromechanical capabilities (i.e. mobility, manipulation and victim detection) or increasing its final price but why we should pay this high price?! Unfortunately I haven’t found the answer in the RC RRL rules yet; even worse, heavier robots sometimes have potentially better performance in RC RRL (note that the robots of iRap-Pro – the first place award winning team of RC09 and RC10 – are definitely heavier than 30 Kg.).
If we accept that the performance of a mobile robot is proportional to its weight, it cannot be right to evaluate a 30 Kg robot with an over 50 Kg. one. Considering the very tight schedule of RoboCup competitions, it won’t be possible to have several weight-class based (e.g. light weight, heavy weight etc.) evaluations. On the other hand having weight-classified evaluations may not encourage participating teams to work on reducing the weight of their robots.

My suggestion is to utilize “dynamic elements” in the arena so that the difficulty of traversing them changes based on the passing weight. An example of such an element is shown in the top of this post. This element is just like a simple 45 deg slope but it is connected to the elevated floor using a pair of sprung joints instead of hinges. Now, the distance between top end of this ramp and the elevated floor will change based on the amount of mass on it. As it’s shown in the picture, the robot with mass “M” will face a gap with length “D” which is larger than the distance “d” caused by the robot with mass “m”. This means that a robot should have greater mobility if it is heavier!

New Scoring Method for RoboCup Rescue

“Time” is the most important limiting factor in an Urban Search And Rescue (USAR) mission. Actually the chance of surviving victims drastically decreases 24 hours after a disaster. In this critical situation, rescue personnel should avoid any risky tasks. Published texts indicate that surviving a trapped rescuer takes about 4 hours for a team of 10 rescuers and it’ll take almost 10 hours if the rescuer is injured or entombed!

Rescue teams can reduce the risk of surviving victims by means of robotics. Robots can provide accurate information about the collapsing walls, hazard materials etc. what rescue dogs cannot do. One may want to put a camera on a rescue dog to inspect collapsed buildings but, these video streams certainly won’t provide the environmental awareness that a smoothly controlled tele-operated robot can. In other words the “accuracy and quality” of environmental awareness (identification of victims can also be considered as a part of environmental awareness) provided by robots is the most significant advantage of these instruments in comparison with rescue dogs. We can also use these two parameters to have a rough estimation about the success of rescue phase in a USAR mission: “the more accurate and usable information one gets in a search phase, the more victims are expected to be survived in rescue time.”

In RoboCup Rescue scenario that is a standardized simulation of real disasters, rescue teams very well know how many people are injured, who they are and where they are located (rescuers may get this information from a survived person in real situations). The task is “to explore the arena by a team of robots with different levels of autonomy in a limited time to gather accurate and reusable information about the surroundings and victims.”
Like in a real USAR mission, here we can expect that a victim will be survived if a safe path to his/her location is discovered. Therefore, a victim should first be detected then his/her exact location be marked on the currently generated map of the environment to start rescuing. If the team cannot find a safe path to the location of detected victim, the rescuers should probably put their own life in a serious danger to survive that victim. This means that a “search phase” is not accomplished unless rescuers can prepare a map of the environment having accurate location of victims with a path for following. If we suppose that the chance of rescuing an identified victim linearly increases with the quality and accuracy of generated map, the following formulas can be used for evaluating general performance of a rescue team:

(1) Score of a detected victim = (mapping score at the point / maximum mapping score) x (victim identification score / maximum score of victim identification)

(2) Final score of a mission = (100 / QTY of all victims) x (sum of victim’s scores)

As an instance, assume that a team could find 5 victims out of 12 in a mission and they could identify all the signs of life (victim identification score = 30) while they could generate a half accurate map (mapping score = 10). Their final score is calculated as the followings:

Score of each detected victim = (10 / 20) x (30 / 30) = 0.5
Final score = (100 / 12) x (5 x 0.5) = 20.8%
This shows that the mission was 20.8% successful!

How would it be if they could find 3 victims with a completely accurate map?
Score of each detected victim = (20 / 20) x (30 / 30) = 1.0
Final score = (100 / 12) x (3 x 1.0) = 25.0%
It seems that they really need to improve their mapping for the next year!

Now let’s take a closer look at this scoring method:
Since final result of each team is calculated by adding all identified victims’ score which itself is the product of mapping and victim identification results, the final score highly depends on efficiency of applied algorithms (Computer Science) and robustness of utilized electro-mechanical systems (Mechatronics). Obviously a team should be very well prepared in terms of mechatronics, computer science and human resources if it wants to be successful in the competitions. Unlike the existing formula, a team will not be able to completely cover its weakness in one of the aforementioned areas by concentrating on its capabilities in the other fields. This can also be considered as a step towards eliminating operator’s skill in success of rescue teams (the end users of rescue robots won’t necessarily have all the capabilities of a well trained operator in RC RRL).

Advantages of proposed method:

• More accurate and still fair: As it was discussed, a team should be the best one in mechatronics, computer science and human resources if it wants to win the place award because these three factors have less effect on each other comparing to the existing formula.

• Easy to understand scoring result: Everyone (competitors and audiences) can easily understand how the performance of a team is even without comparing it with other teams’ scores when it is presented in percentage.

• Flexibility: Since the final score is a fraction of maximum available score, it does not depend on how many victims are placed in the arena. Therefore, the committee can freely add or remove some victims in an event while the performance of participants will be easily evaluated with previous events.

• Encouraging development of accurate real-time SLAM algorithms: Generally accuracy and quality of mapping algorithms reduce when the speed of a mobile platform increases. So, teams need to develop their mapping algorithms to have more accurate maps when they drive their robots faster to find more victims.

• Necessity of search strategy: One may want to find more victims to have grater score while another one may decide to get complete score of each victim. So, having a search strategy based on technical capabilities besides compromising between finding more victims and having more accurate maps will be inevitable.

• Shortening the score gap between auto-based and tele-based teams: Most auto-based teams produce very high quality maps. So, their mapping coefficient will be greater than tele-based teams. As a result, they can probably get more score if they find a victim.

Disadvantages:

• Map scoring: The presented scoring formula highly depends on mapping score. Since the automatic map scoring system is still under development by the RC RRL committee, final results will rely on human factor till a reliable automatic map scoring system becomes available.

However, there may be several teams that are not interested in AI problems or mechatronics. Those teams can always try their chance at best in class challenges.
At the end I think the proposed scoring formula can help to improve the quality of our highly valuable RoboCup Rescue Robot League.

A brief history of robotic mapping in RoboCup Rescue

At the very beginning of RoboCup Rescue, teams usually provided a hand-drawn map of the explored area to indicate the location of detected victims after their missions. Although these hand-drawn maps may be quite enough for rescuers to find the victims but, it didn’t have any technical challenge for academic side. In 2006 the RoboCup Rescue committee decided to restrict acceptable mapping methods to the automatically generated grid mappings. Although several teams (mostly European ones) were developing SLAM algorithms even before 2006, they didn’t have a similar method to represent their mapping results. Therefore the committee standardized the mapping format to be able to quantify the accuracy and quality of these maps. With these standardized mapping formats, now the committee members are working to develop an automatic map scoring system.

How can team leaders help RobCup Rescue?

Every year after coming back from RoboCup competitions, I send my ideas to the mailing list of RoboCup Rescue Robot League hopefully to improve the quality of our league. Unfortunately I have been one of the few people (if not the only one) who shares his/her ideas in the mailing list during these years. In fact the Rescue Robot community of RoboCup competitions is not that much active and we need more collaboration from team leaders in this area.
As far as I know, it has always been a big challenge for the organizing committee (especially Adam) to make the league closer and closer to the real scenarios while it is standard and of course interesting for the audience. In this way, they should consider several diverse issues (i.g. First Responders’ technical requirements, the existing technologies, standard test methods and technical level of participating teams). Obviously the Executive and Technical Committees will reach to a better conclusion if they know the ideas of participating teams (at least about technical issues).