Dear all,
In our team, we discussed about our simulation improvement. We know that lots of difference can be mentioned between our simulation and real world, but we should make progress step by step. So we chose some of our discussion topics which are listed below:
1 - Perception:
All the things which have been mentioned before are nice challenges, but we'd like to add some more things.
First to eliminate the sensible distance(and also hearing distance), in other words changing it to infinity and defining a function to produce chance of perception. The farther the object is the chance of perception is reduced more.
Second about fire perception: if the fiery building could be sensed(not because of its fire), the agent sense it as a building. But if a fire is sensed just because of its smoke, the related building should not be sensed and just an approximate fire point (depending on distance to fire) should be sent to agent.
Third to have time (day and night) in simulation which can affect the perception power. For example if the rescue job is being done in sunset the vision power should be reduced little by little.
Forth KA_SENSE and KA_HEAR packages problem: Suppose that an agent moves from point A to B in a cycle. What does this agent sense? In our simulation the area in sensible distance of A and B is just sensed, but the agent has passed a route and it should have also seen and heard anything along its route even fires. The sensing and hearing chance while moving, should be less than the one while they are not moving. Therefore, the KA_SENSE and KA_HEAR packages should also contain the information about the agent's route.
Fifth about sensing agent's action. It is really a good idea. For example, a fire brigade submits an extinguish action, and kernel does not receive this package. How can this fire brigade understand whether it has done its own action or not.
2 - Speed:
If the new field, speed, is added to move action, it will be more natural, just like water power in extinguish action. Think about an agent which wants to search an area. Does it always move with maximum speed. Maybe it passes away and because of its high speed it doesn't sense the area around its way. So it's better that the agent moves with adjustable speed.
3 - New flying agent:
Helicopter is a necessary device in real world rescue missions. If we define a new flying agent (or maybe new capability for previous agents) we'll have a more complete simulation environment. Think about this agent while flying above unreachable area and reports what's going on there. It also can do other actions such as transferring civilians to refuge. If this idea is accepted we can present a more accurate pattern of this agent.
4 - Traffic in buildings:
It seems to be a nice feature that capacity of buildings be limited and we support this idea. But eliminating the extinguish ability from other buildings is not a good idea, because if agent can go inside the building, it can also go on the roof of the building and extinguish other buildings. And one more point for fire brigades traffic problem, the nozzle position if differs from the agent's position, the agent is more capable and can extinguish farther positions. It reduces the traffic jam.
5 - Changing the movement order during a time-step:
If we take a closer look at this move-related problem, we realize that it is not only for move action, and all of the actions suffer from same problem. For example if a fire brigade decides to extinguish a building with water power 1000, and if it changes its decision while this action is going on, it can not do anything and 1000 units of its water is wasted. We think that this problem refers to long time-step we have, making decision once a minute. If we reduce the time-step, which is not possible now and will be good for future years and better hardware, this problem will be disappeared little by little.
6 - Reducing the ability of agents while they are damaged:
It is unrealistic that a damaged agent works as well as an intact one. For example reducing the maximum speed of move depending on agent's hp and damage.
7 - Building type:
We have 3 types wooden, steel and concrete. Adding some new features seems to be nice. For instance, explosive buildings like gas stations, self-extinguishing buildings which have fire extinguishing system inside, etc and all of them can be wooden or other types.
8 - Flammable refuge and center:
Refuge is supposed to be a safe place and should not be damaged easily. If we want to make it flammable the chance of ignition should be really low, approximately zero. What's more, adding this feature makes the rescue job a little nonsense. Centers also can be self-extinguishing places to avoid communication problem, because they are really important places.
9 - More fires:
If we take a look at real earthquakes, we rarely encounter a big fire (for example half of a city burned in fire). But it's a very simple thing in our simulation even for best teams to lose a big area in fire. In real disaster rescuing buried people is the most challenging job, and coping with a big fire happens hardly ever. 8 fire from approximately 800 buildings at the beginning (about 1%) is really a huge number. If we map it to a real city with 100,000 buildings, the unrealistic aspect of it will show itself better, 1000 fire at the beginning. So instead of increasing number of fires, making fire perception a little harder and more real which is mentioned in second perception unit at the top, seems to be a better idea.
Regards,
Hamid Hamraz & Shams Feizabadi,
Caspian members,
Iran University of Science and Technology
---------------------------------
Do you Yahoo!?
Discover all that’s new in My Yahoo!
This archive was generated by hypermail 2.1.3 : Thu 18 Nov 2004 - 14:44:59 GMT