March 2002 Jrobots Challenge: Results

To read the complete results of the last contest go to the Challenges page.
Here some comments follow:

During the last contest we experimented the effect of the Virtual Clock Generator on the on-line arena.
As you know, the machine dependency of Jrobots was a great problem in the past months: A lot of players complaint because their robots didn't work as they expected in the on-line arena and the winning percentages were always lower than the percentages they got off-line during the test phase.

Finally, it seems that a solution was found and I want to thank again Alan Lund and Walter Nisticò for planning the new clock that beates the time of the matches and for the code they wrote, but...
yes, there is a but: In fact, we couldn't expect that all the robots in the arena were going to react well to the virtual time. Reading the results of the latest challenge, we see that Mouth gained a placing in Single Mode and becomes the best robots in Team Mode with a convincing gap from the second placed.

The new clock gives to the robots more time to think and no more blackouts occur when the CPU is busy in other operations. The Team Mode shows clearly that the native thread scheduler gave very few time to each robot, while now all the 32 robots has enough time to think, slowing down the simulation, of course. Well, designing a Team Match with four teams of eight robots each was a poor choice and the virtual clock demonstates it with no doubts.

Slightly different behaviours apart, now we can be sure that the arena works fine in a wide range of conditions, or at least better than before.
The alternative solution to the clock generator was a interpreter of Java bytecode and I'm trying to write it for JASTROBOTS. As you remember, the first version of the "JVM written in Java" was one thousand times slower than native Java: a very disappointing result. Now I've implemented a number of memory hungry optimizations and the highest speed I've reached is only 25 times slower (the test code was a cycle filled with arithmetic operations on integers, float and doubles). I think it is a good result, so this solution seems to be slower, obviously, but it is also feasible.

This month the defeated robots are: IlTristoSmorzatore (an old robot strong in Single Mode but very weak in other kind of matches), 1070bytes (the smallest effective robot around), OneEyedWilly (the ancestor of OneEyedWillyII), Parasite (that belonged to the group of the Germ-series) and, finally, the novices Malaysian and Cougar. Only 9 robots are now in the arena!
The winner of single and double category is still IonStrom and the virtual clock generator effect moved it to the second place in Team Mode, while Germ and GermPlus suffered it and lost places.

Using the Alan Lund's Watch utility (that you can download from the Files section of the mailing list), we can foresee that only 3 robots are going to survive the next challenge, if nothing new happens: They are KillerBees, Mouth and IonStrom. All the other robots can't gain the requested percentages to survive. The foresight considers a lower number of matches respect to a normal tournament (744/789/207) and I hope that the reality is going to be not so destructive.
If we run out of robots, I'm planning to organize a several months long monster summer challenge, including in the online arena all the best versions of the 131 robots that partecipated to the Jrobots Challenges in the past two years. I know it's a crazy idea, maybe even not useful or impracticable, but I'm sure that some old robot will perform better using the new simulator and I am curious to see a Jrobots all-times placing list ;-)

Now you can upload your robots to win the May 2002 Jrobots Challenge (May 4^th-25^th 2002) aka Only One Is Going To Survive!

• The April 2002 Jrobots Challenge starts now.
  Remember that you can send your robots even in the course of the challenge, so you're never too late. The latest robots are stored in the file jjrobots_challengers.zip. You can download it to test off-line the behaviour of your robot against the other challengers. To read the results in real-time simply visit the Battle Applet page.

• Remember also that the current challenge use a virtual clock generator to beat the thinking time of the robots, so you'd better to test your old robots with the new SDK and possibly modify them to get better performances.