This mod for the HBS BattleTech attempts to make the AI a bit smarter.
This mod requires [https://github.com/iceraptor/IRBTModUtils/]. Grab the latest release of IRBTModUtils and extract it in your Mods/ directory alongside of this mod.
Jumping Heat change
The following values can be tweaked in mod.json to customize your experience:
- Debug: If true, detailed information will be printed in the logs.
Part of the problem is that BT has too many options. These options have to be encoded into decision tree calculations, which is a large series of branching yes/no questions. What's your role (sniper, brawler, escort, etc), what movement best helps that purpose (from a choice of 20-60 hexes and 2-3 movement types (move, sprint, jump)). Balance that against which position makes the most sense when you have 4 weapon bands, and heat to balance as well.
There are many things that could help that outcome - running different movement calculations in parallel, making a stronger determination of unit purpose before going through the tree logic to determine specific actions (which is present but not heavily used), pre-calculating weapon fire from each position independently, etc. Short-cutting the weapon calculations by applying a mean value to the shots * projectiles * damage calcuation, instead of doing each one one by one, etc.
I've been trying for a while to get a profiler hooked up to confirm my suspicions, but I think three changes would make a major impact on performance:
- Start running the decision tree in the previous actor's turn. A suboptimal decision that doesn't take into account the last action but executed immediately probably increases player satisfaction immensely.
- Normalize weapons and cluster them from a single emitter, instead of resolving the damage weapon-by-shot-by-projectile
- Implement stronger 'lance commander' logic (already in AI) that defines what the lance wants to do, instead of letting it waffle actor by actor. Pick a model on the player's side to just punish, instead of letting each actor do their own thing.
AIUtil uses: if (AIUtil.Get2DDistanceBetweenVector3s(sampledPathNodes[i].Position, this.unit.CurrentPosition) >= 1f)
But add Mech uses:
public void OnJumpComplete(Vector3 finalPosition, Quaternion finalHeading, int sequenceUID)
{
float num = Vector3.Distance(base.PreviousPosition, finalPosition);
this.AddJumpHeat(num);
I wonder if it's a difference in 2D vs. 3D vector calculation That actually would explain it fairly neatly I bet they don't overheat on a flat plain But only when they are jumping vertical distances I bet that's it, yeah.
- Add crit-seeking potential to weapons
- Aggregate weapons into clusters
- Clusters should reduce their count to work with overheat calculations effectively
- Calculate firepower loss from friendly unit shutdown
- Calculate melee retaliation properly
- Prevent melee from light units that won't benefit
- Incorporate CAC AoE and inferno effects
AI logic invoked by AITeam.TurnActorProcessActivation
- Checks for opfor reservation
- Checks for reservation pre-reqs
- Assignes role to unit
AITeam.OnUpdate() -> AITeam.think()
Sequence starts at ShootAtHighestPriorityEnemyNode:Tick() ExecuteStationaryAttackNode.Tick(). Both invoke AttackEvaluator.MakeAttackOrder()
- Iterates over designed targets for lance, making an attack order for the target. Evaluates the firepower reduction from the attack
- Then iterates over every enemy unit that's not dead, and is an abstract actor. MakesAnAttack order, and evals firepower reduction from attack. ** For each target, evaluates additional damage from friends attacking an evasive target (because of evasion strip) ** If damage is greater than BehaviorVariableName.Float_OpportunityFireExceedsDesignatedTargetFirepowerTakeawayByPercentage, chooses opportunity fire *** If opportunity fire, tries to create a MultiTargetAttackOrder if possible.
MakeAttackOrder invokes MakeAttackOrderForTarget, and evaluates FirepowerReduction from attack. MakeAttackOrderForTarget is only ever invoked by AttackEvalutor.MakeAttackOder