Bullies and Dweebs
In this simple pursuit-evasion game, a dweeb will look for the closest bully and take a step away. A bully will find the closest dweeb and take a step towards. Interesting emerging properties are bully teamwork, bullies cornering dweebs, dweeb flocking and dynamic Voronoi partitioning. Bully teamwork, though, has its price since bullies combine & loose effectiveness as a herd. Dweebs can also combine - but twiddle can let them separate again
- No Dweeb Tiddle
- Dweeb Twiddle (More Interesting!)
- One Bully - Many Dweeb
- Drunk Bullies
- Frenzied Bully Mobs (Most Interesting!)
- Paralyzed Bullies
- Diagonal Run
- Self Sacrifice
- Dweebs Playing Possum
Dweebs adhere strictly to taking a step from the nearest bully - to the disadvantage of the dweebs. The massacre effect emerges. There are two bullies and 5,000 dweebs in this emulation, and, in the end, only 9 death spots. Dweeb deaths as a function of time, display a staircase form - indicative of recurring massacres.More Information
Dweebs occasionally take a random step. This is great for dweebs escaping from walls and corners - but also allows a dweeb to back into a pursuing bully. There are 25 bullies and 2,000 dweebs at the start. The bullies don't even get half the dweebs after 750 iterations.More Information
Watch how a bully tags at least one dweeb at a time. Dweeb twiddle is used. Occasional massacres.More Information
Adding twiddle to the bullies makes them look drunk.More Information
Adding twiddle to many bullies ,amazingly, aids in their collective capture ability - even though one at at a time the bullies look drunk. This is a fascinating emerging property wherein inexactness on the bully's part actually aids the collective in the capture. The collection of the bullies resembles a frenzied mob. Very fascinating viewing!More Information
when the Bullies are paralyzed, the dweebs retreat to the Voronoi partition.More Information
A number of student programmers wrote bullies and dweebs programs. Suggested rules for the dweebs included (1) stay away from walls and corners, (2) move away from the nearest bully, (3) avoid the nearest dweeb under pursuit. Other rules could be applied. No further description of the fuzzy dweeb interactions were given. Each programmer implemented their own disjunctive logic control for the swarm using different implementation details. The swarms were evolved to maximize the median life of the swarm by changing the fuzzy control variables. Only the dweebs changed behaviour. The evolutionary algorithm, in all cases, was the particle swarm optimization (PSO). Here are some to the interesting emergent behaviours that resulted.
Here, dweebs sacrifices themselves as a crowd of other dweebs watch in the corners.
The emergent behavior is that a single dweeb will sacrifice itself to distract the bullies, so the main group of dweebs have time to get away. This is evident in the video at 0:30 and 0:50. A lesser emergent behavior is that the bullies tend to form a pincer to corner groups of dweebs.
In this simulation, the bullies were set to aggressively attack the dweebs. In order to accomplish this, they would anticipate a dweebs step away from them and proceed to move there. This means when a dweeb was one step away, the bullys next position change would move it two steps: one for where the dweeb is now and one more for where the dweeb would move. Using this strategy, the bullies killed the dweebs efficiently. That is until I optimized the dweeb's half life. Now the dweebs have a good strategy too. They play dead. When a bully is really close, usually within one step of it, the dweeb stops moving and pretends to be dead. This causes the bully to jump right over it, since the bully is anticipating the dweeb to move.
The optimized vector found by Particle Swarm Optimization, takes roughly 20% to 25% of the fuzzified closest infected dweeb and closest wall distances, the momentum value, and the twiddle value, with the largest weight being the twiddle. An extremely small weight for the distance to the closest bully was found, which makes sense, because when a bully is right next to it, the dweeb doesnt move. This is why it looks like it is playing dead.
- Animation of how ants use pheromone to get food back to their nest. Click on the animation to drop food. Click Pheromone to see the pheremione stregth. Add ants by ckicking of the lowe rleft. Be prepared to spend your day watching.
- National Geographic article on swarm intelligence.
- Swarming on the Battlefield (Google books)
- Swarm Intelligence in Business Harvard Business Review