Page MenuHomeFeedback Tracker

Some possible discrepancies with regards to AI accuracy with small arms
New, NormalPublic


Even though I have almost no knowledge on the inner working of ARMA, I believe I've been able to pin point some anomalies with AI accuracy that may have lead to... Some counterintuitive results in gameplay.

I'll start with what appears to be specific cases of straight oversight. SPAR-17, with otherwise very similar physical performance to other 7.62mm DMR, is configured in cfgweapons to have

aiDispersionCoefX = 6
aiDispersionCoefY = 6;

Which is in line with small caliber assault rifle in game rather than DMRs that are otherwise universally configured to have

aiDispersionCoefX = 2
aiDispersionCoefY = 3;

If I interpreted with any correctness that's 6x larger dispersion area than otherwise comparable weapons.

MXM/CMR-76 also suffers from similar inconsistency. Where as the former's aiDispersion closely matches the MX assault rifle(4*6), the latter matches that of a marksman rifle(2*3).

The worse offender in this regard though may be the machineguns. Since they are configured universally to the same standard I'll assume it is all intentional. But with aiDispersion configured to be on the order of 21*24 the dispersion area can be more than a dozen to dozens times of other classes of weapons. Adding to it is the AI's preference to always fire in bursts, the firemode which is further configured to have aiDispersion of 2*3 than 1.4*1.7 of single shot. On top of which the shots subsequent to the first within the burst almost never contribute to the lethality at normal combat range, only serves to increase the interval between aimed shots. While I could get behind the idea that machinegun in AI's hand should emulate their real life battlefield role, the non-existence of actual threat really diminishes it's ability to do so. It's all the more ironic that machinegun in various flavors are probably the most effective smallarms in the hands of players, especially since the introduction of resting/bipod feature that AI can't make use of.

I'd go on to suggest that AI seemed to have the appropriate level of lethality in the earlier days of ARMA. When dropped to prone stance they seemed to be able to compensate for recoil better, recenter between bursts faster, and generally steady aim better. How exactly it worked is unclear to me, but within the current mechanical limitation a prone bonus of some kind could serve as an abstraction of the resting feature for players. Some fire discipline change could also be helpful, on top of addressing the artificial inaccuracy with the weapon class itself.


Operating System
Windows 10
AI Aiming / Shooting
Steps To Reproduce

Due to my limited knowledge, my methodology involves simple "turkey shoot" scenario, gauging the time for AI to eliminate a specific set of non retaliatory targets. But even with my rudimentary test I would argue the result is very conclusive. Weapons configured with large aiDispersion consistently underperform in the hands of AI relative to weapons with otherwise similar characteristics. Machineguns routinely take several times longer or more to engage all targets.

Additional Information

Upon closer examination I will acknowledge that at more extreme AI skill/difficulty settings the accuracy does start to equalize, possibly due to their effects on aiDispersion I assume. It might explain why the hit probability was never differ by exactly the number of times of the aiDispersion difference. Still at the most normal difficulty/skill level, and especially the lower ones the effect is quite pronounced.

Some earlier discussions on the forum

Event Timeline

slicedice created this task.Feb 9 2019, 2:34 AM