Melee-Specific Targeting

Origins

Melee-specific targeting was pioneered by ABC, whose bot Shadow set the standard for evaluating every enemy on the field at once instead of locking onto a single target, and was further documented by the RoboWiki community.

GuessFactor Targeting works because an enemy dodging this bot's bullets tends to move a certain way relative to this bot. In melee, an enemy is rarely dodging any one bot in particular, it is just trying to survive everyone at once, so that relationship weakens. A gun built around one assumption from Simple Targeting and the statistical pages needs new judgment calls before it works in a crowd.

Why GuessFactor loses its edge

GuessFactor Targeting leans on the enemy moving roughly perpendicular to the bot doing the shooting, since that is what produces a clean escape-angle distribution to learn from. In melee, most bots spend their movement budget reacting to several threats at once, not specifically dodging this bot, so the perpendicular pattern a guess-factor gun depends on shows up far less often. The technique still works, it is just a smaller edge than it is in a duel, and a bot that leans on it exclusively is leaving a duel-only advantage on the table.

Distance decides the algorithm

Close enemies favor Linear Targeting, since a bullet that only has to travel a short distance does not need much lead to land. Far enemies favor Head-On Targeting, a simpler baseline that holds up once distance shrinks the payoff of tracking angular movement precisely. A gun that blends the two by distance, weighting linear more heavily up close and head-on more at range, often called percent-of-linear, beats committing to either one alone.

Scans arrive late and uneven

A Melee Radar sweeps past every enemy in turn instead of locking one, so the gun sees each enemy roughly every 8 ticks on average instead of every tick. That gap breaks Pattern Matching, which expects a steady log of positions, unless it is rewritten to tolerate uneven gaps. Statistical guns cope by interpolating a missed scan linearly between the two real ones instead of discarding the gap, and by reading the sign of an enemy's velocity: if it flips from the previous scan, that enemy is probably accelerating, a cheap signal even from sparse data.

One gun, every enemy

Shadow's melee gun changed the question from "where will this enemy be" to "which enemy can this bot actually hit." It computes a firing solution for every enemy still alive, not just the closest one, weighs each solution by hit probability, roughly the inverse of the distance to that enemy, and fires at whichever solution scores highest. The reported payoff was about a 10% hit rate increase over locking onto one target and hoping.

That framing opens two judgment calls a single-target gun never has to make. Preying on the weak, favoring a low-energy enemy's solution over a healthier one even at slightly lower hit probability, claims kills instead of just damage. Checking a miss's flight path catches a shot that would fly past its target and waste itself on a wall instead of a second enemy standing behind it.

A melee gun scores a firing solution for every enemy present and fires at the strongest one.

Further Reading

• Melee Strategy - RoboWiki (classic Robocode)
• Shadow/Melee Gun - RoboWiki (classic Robocode)
• Segmentation - RoboWiki (classic Robocode)