Using AI To Train Firefighters

Roland Piquepaille writes "Computer scientists at the University of Southern California (USC) have developed DEFACTO, a training program which uses artificial intelligence (AI) to help firefighters practice simulated emergency situations. The system is currently used by the Los Angeles Fire Department. DEFACTO has committees of AI 'agents' which can create disaster scenarios with images and maps seen in 3-D by the trainees. The software agents also evaluate the trainees' answers and help them to take better decisions. As one LAFD captain said, 'You can see if you're heading toward a mistake much more quickly.' Read more for additional details about this AI project and a photo of a LAFD Fire Captain using the system."

Comments

If you ask me...

[RuBLed]

I believe the fire captain is playing.. errr.. beta testing Sim City Societies, I swear I could see the Firefighter dispatch tool...

"America's Firemen"?

[weighn]

so can the USC use some of this code as the basis for a game? This could be used as both a recruiting tool and an aid to help recovering pyrophobics.

I'm not suggesting that it would be a shame for all this hard work to go into just training guys to save lives and property :) but the screenshots suggest that there could be elements of strategy (which block to approach from, consideration of wind conditions) and some cool FPS with the water cannon. Possibly some riot quelling action too ... are there any firefighting sims out there? Perhaps a multi-player which pits firefighters against arsonists?

Questionable Results

[Vskye]

First off, it's a AI program. This will not give you that much practical experience compared to being in a more traditional controlled test environment. I used to run ship fire simulations, with actual fire and smoke in a building setup for this type of deal when I was in the Navy. We actually trained a group of Iranians back in the day, and I remember several trying to get out of the port holes on the building, because they freaked out. (ever see a 2 1/2" hose flying around? this will put a serious hurt on your team if nailed.) I guess my whole point is that actually going into a test environment on a moments notice, donning your equipment and getting a first hand view of smoke/fire is a better experience, IMHO.

Re:Questionable Results

[hcdejong]

As others have said, this simulator isn't meant to replace basic training. It's about planning how to attack a major fire.
 
...a more traditional controlled test environment.

The problem with such an environment is that it's limited in size, and has a fixed layout. This is no problem for initial training, but experienced firefighters will go in knowing what to expect, which skews the results.

With the simulator, you can present scenarios that are guaranteed to be unfamiliar to the trainee, and on a much larger scope than physically feasible.

Similar systems are used in the army: they have lots of hands-on weapons training for the soldiers, but also "command post exercises" where the trainee commanders just shuffle paper, issue radio orders etc. and no bullets are involved. In such an exercise the commander can learn how to manage his division, in scenarios that can't be played out in real life (e.g. because there's no training ground big enough to do battles with an entire division, plus there's the cost factor).

Re:Questionable Results

[$RANDOMLUSER]

The thing that I'm wondering after reading TFA is whether or not the system gives the trainees too much information. The real skill in these kind of situations is making good decisions on incomplete data. I hope this training approach doesn't teach people to rely on information they won't have in the real world.

Re:Wrong focus.

[lunartik]

It says it helps them improve decision-making, which is what disaster scenarios are about.

I used to write and run crisis training for a large multi-national, and we expected that the participants would make mistakes. The basic thing we tried to do was to give them a chaotic situation and to teach and re-enforce their roles and responsibilities. The main thing was that they knew what their role was, and that they stuck to it. All crisis situations are basically getting something under control. It is a bit different for first responders, but not much.

In our case it was a bit different because we were dealing with people who had a role within the corporate crisis structure, but it was not their actual job within the organization (normally they might head legal, HR or finance - for example). So we had to make sure they were periodically trained so that they would be familiar with how everything worked in the event that the crisis mechanism was activated.

Training for firefighters is a bit different because it is their main job, and it is reinforcing skills they will use regularly.

The land of kangaroos and drop bears

[mazzanet]

Over here in the lovable land of .au, we (CFA) and other fire agencies around the country already use a training simulator made by VectorCommand in the UK (link) which is very similar to this package in the OP. Career staff particularly, as well as volunteers, are readily ran through various exercises and drills on it. One of the most useful features is that it a scenario can be run individually or with a team of users (with networked computers).

Different Approach

[Raptoer]

This is not to replace real training at all, I totally agree that this type of training can and never will be able to (in the foreseeable future) match a real training simulator. This is more of management type training, where a firefighter learns how to make decisions with disaster type situations.

The real advantage of this system of the previous version is it no longer ties up an entire fire team in order to run the simulation, it comes up with truly random situations (a person is more likely to train people about things they experience all the time, well you also need to be able to meet unexpected situations), and it is easier to run. The changed graphics is really just a different UI, it is not nearly good enough to trick someone in to thinking it is the real thing, but as long as you're making the same type of decisions under some pressure, the graphics do not matter. The increased graphics might make it easier to relate to a real situation. An example being a red box is a firetruck vs a more complete model, it just helps with the visualizations.

As a fire officer myself -- this looks limited

[CFD339]

I'm a Lieutenant on a small department, and have been part of many "simulator" drills using not so artificial intelligence. We use an overhead project, software that allows drag and drop visual and audio changes to simulate the progression of a fire against a background image, and the real intelligence of a senior officer running the drill. The purpose of the drill is to give experience to the firefighters and officers making decision on how to attack the fire and when to take certain actions.

It is NOT about the individual firefighter on the line doing his job -- he's not meant to be focused on the big picture. He's got to focus on his local task as assigned -- search, vent, attack, etc.

The important thing, is that the person running the drill has a situation in his head about what caused the fire and how it will progress through the building, and can adjust the progression as a result of the decisions made by the officer making decisions. Once done, a great benefit of the experience is the discussion. Another huge benefit is the practice at simply making best use of radio traffic and keeping the situation in mind all at once.

The only advantage I see to this software they're showing is that it requires less people so you could have more people being in charge of the scenario and learning. They lose the experience coordinating the radio traffic, and the discussion at the end of how their tactics impacted the results would not benefit as many people.