Here's my one minute postmortem of Arena:

What went right
1) The concept is interesting. Who doesn't like blowing up robots?
2) Added several features to engine.
3) Spectator mode with multiple camera angles and a sports-like leader board allowed non-scripters to enjoy the project.
4) KISS. There where a ton of things I wanted to add (triggers, jumps, power-ups, etc.), but I stuck with the core. It's hard enough to track and kill a single robot without worrying about pit-traps.

5) Limiting scripts. I figured out quickly that there had to be rules on what you could and couldn't use with scripts. Being able to teleport or just remove() robots would have been no fun.
What went wrong
1) "Scripting is hard." Few users in 2002 could write scripts. Part of the reason for the Arena project was to encourage people to learn scripting, but only a couple people tried.
2) "AI is hard." Users either fear AI, or think that they can just use a bunch of random numbers to beat their opponents. The best scripts where the most simple ones.
3) "Arena is hard." Writing a robot from scratch required a certain frame of thought. Once you had the basics down, tweaking a robot wasn't too difficult but getting something started was frustrating.
What I would do for Arena2:
1) DATA-DRIVE! Do not require new users to learn scripting, and do not force weird limits on experienced scripters. Instead, make a simple command language (like LOGO) and allow users to write using that.
2) Use Physics. Now that all versions have physics, use it.