RealTime Target Selection and Battle Outcome Estimation in ORTS

1
Real-Time Target Selection and Battle Outcome
Estimation in ORTS

• Doug Demyen

2
Outline

• Introduction
• Problem
• Other approaches
• Observations
• Targeting Issues
• Sub-battle Optimality
• Proposed Approach

• Solution Representation
• Unit Attributes
• Simulation Environment
• Selecting Target Order
• Running the Simulation
• Restarting the Algorithm
• Conclusion

3
Introduction

• Outcomes of battles in RTS games can vary greatly
  simply based on unit target selection
• Want to free the user from such tedious tasks
• But need to work effectively in the constraints
  of real-time

4
Problem Definition

• Proper strategies such as target selection can
  win or lose a firefight in an RTS game
• For example, concentrating fire to eliminate
  enemy units one at a time can be effective
• Requiring the user to perform such
  micro-management detracts from high-level tactics
• Problems such as over-concentrating fire and
  selecting the units to target first, must be
  addressed

5
Previous Work

• Attempts have been made to solve a battle for
  the optimal actions of the units involved
• There are two main drawbacks to this
• First, this can take a lot of time and resources,
  both of which are scarce in a game environment
• Second, these approaches lack the flexibility
  needed for events that could change a battle
• For example, units could join either side or a
  bomb could destroy or weaken existing units

6
Observations

• In an RTS environment, an optimal solution may
  have to be traded off for an effective one
  because of constraints
• To simplify the search for such near-optimal
  solutions to work within these constraints, one
  can make the following observations

7
Targeting Issues

• Concentrating fire can be done to excess and may
  result in wasted attacks
• This is a small drawback as the loss is often
  negligible
• Thus for the purposes of this approach, each unit
  will only change targets when the current target
  is destroyed
• This is a great advantage since each unit doesnt
  have to decide what to do each frame

8
Sub-Battle Optimality

• No matter the conditions of the battle, the
  strategy is relatively constant
• The goal remains to inflict the most damage and
  casualties and sustain the least
• Thus targeting decisions can be made for each few
  seconds instead of all at the start
• Also, targets can be assigned irrespective of
  future events, at which point they can be
  reselected, providing more flexibility

9
Proposed Approach

• The application of genetic algorithms will
  provide
• Fast convergence on an optimal solution
• Flexibility to abort with a solution if time has
  run out or resources are needed

10
Solution Representation

• From above, targets are selected for each few
  seconds (say 2 5) at a time
• Let k be the most units any unit is likely to
  destroy in that time and n1, n2 be the number of
  units for armies 1 and 2, respectively
• Thus the targets for army 1 for this time consist
  of k n1 integers from 1, . . . , n2

Unit 1
Unit 2
Unit n1
Example k 3
11
Unit Attributes

• This approach will be mainly concerned with the
  following attributes of different units
• Hit Points the damage the unit can sustain
  before being destroyed
• Attack Strength the damage the unit can inflict
  in one time step
• Recoil Time the number of time steps required
  between each attack
• (Attack range and speed are generalized into
  whether one unit can target another)

12
Simulation Environment

• A simulation is run with the units represented
  using the above attributes and using time steps
  representing a few frames in real-time
• Targets are selected for each unit and the
  simulation is played out in time steps
• Each time, step, each unit deals its Attack
  Strength in damage to its target, having to pause
  for its Recoil Time
• When a units Hit Points fall below zero, it is
  removed and its attackers aim at their next
  targets

13
Target Selection

• The k n1 targets for army 1 for the length of
  the simulation will be encoded into a genome to
  be used by the genetic algorithm
• The most effective target order will be found
  using the evolution of the best outcomes
• These will be determined by the amount of damage
  and casualties sustained by both armies

14
Running the Simulation

• When the algorithm is run, a generation of
  genomes is created at random or with a simple
  function
• For each genome, the simulation is run and the
  outcome used to determine the fitness
• The best genomes combine into the next generation
  of the algorithm
• After a given number of iterations (or when the
  algorithm is interrupted because of processing
  constraints), the algorithm will return the best
  target- selection genome of the current
  generation

15
Restarting the Algorithm

• Since the algorithm is run only for a few seconds
  of real time and the battle environment often
  changes, the algorithm will have to be rerun
  under a number of conditions
• A unit eliminates all opposing units from its
  target queue
• A unit is destroyed by a source outside the
  battle
• Any new units join the battle
• A unit can target an opposing unit it couldnt
  previously
• A unit cant target an opposing unit it could
  previously
• The time represented in the simulation has expired

16
Conclusion

• This technique should be useful in determining
  effective if not optimal attack patterns
• It is flexible enough to deal with a changing
  battle environment
• It is fast enough to work in an interactive
  setting
• It can also trade off solution quality for
  processing resources as needed

17
Possible Extensions

• This technique could be used to estimate the
  outcome of an entire battle given the following
• What units will join the battle and when
• What units can be targeted by any given unit and
  when
• Outside forces that could significantly affect
  the battle
• Each units targets could be determined, and from
  that the outcomes of each few seconds, and
  finally the battle as a whole
• This could affect whether armies are willing to
  engage in the battle, how the individual units
  would behave, or what future strategies might be
  used