|
|
//========= Copyright Valve Corporation, All rights reserved. ============//
//
// Purpose:
//
// $NoKeywords: $
//=============================================================================//
#include "cbase.h"
#include "ai_movesolver.h"
#include "ndebugoverlay.h"
// memdbgon must be the last include file in a .cpp file!!!
#include "tier0/memdbgon.h"
//-----------------------------------------------------------------------------
inline float V_round( float f ) { return (float)( (int)( f + 0.5 ) ); }
//-----------------------------------------------------------------------------
// CAI_MoveSolver
//-----------------------------------------------------------------------------
// The epsilon used by the solver
const float AIMS_EPS = 0.01;
//-----------------------------------------------------------------------------
// Visualization
//-----------------------------------------------------------------------------
void CAI_MoveSolver::VisualizeRegulations( const Vector& origin ) { if ( m_Regulations.Count() ) { CAI_MoveSuggestions regulations; regulations.AddVectorToTail( m_Regulations ); NormalizeSuggestions( ®ulations[0], (®ulations[0]) + regulations.Count() );
Vector side1, mid, side2; for (int i = regulations.Count(); --i >= 0; ) { // Compute the positions of the angles...
float flMinAngle = regulations[i].arc.center - regulations[i].arc.span * 0.5f; float flMaxAngle = regulations[i].arc.center + regulations[i].arc.span * 0.5f;
side1 = UTIL_YawToVector( flMinAngle ); side2 = UTIL_YawToVector( flMaxAngle ); mid = UTIL_YawToVector( regulations[i].arc.center );
// Stronger weighted ones are bigger
if ( regulations[i].weight < 0 ) { float flLength = 10 + 40 * ( regulations[i].weight * -1.0); side1 *= flLength; side2 *= flLength; mid *= flLength;
side1 += origin; side2 += origin; mid += origin;
NDebugOverlay::Triangle(origin, mid, side1, 255, 0, 0, 48, true, 0.1f ); NDebugOverlay::Triangle(origin, side2, mid, 255, 0, 0, 48, true, 0.1f ); } } } }
//-------------------------------------
// Purpose: The actual solver function. Reweights according to type and sums
// all the suggestions, identifying the best.
//-------------------------------------
bool CAI_MoveSolver::Solve( const AI_MoveSuggestion_t *pSuggestions, int nSuggestions, AI_MoveSolution_t *pResult) { //---------------------------------
//
// Quick out
//
if ( !nSuggestions ) return false;
if ( nSuggestions == 1 && m_Regulations.Count() == 0 && pSuggestions->type == AIMST_MOVE ) { pResult->dir = pSuggestions->arc.center; return true; }
//---------------------------------
//
// Setup
//
CAI_MoveSuggestions suggestions;
suggestions.EnsureCapacity( m_Regulations.Count() + nSuggestions );
suggestions.CopyArray( pSuggestions, nSuggestions); suggestions.AddVectorToTail( m_Regulations );
// Initialize the solver
const int NUM_SOLUTIONS = 120; const int SOLUTION_ANG = 360 / NUM_SOLUTIONS;
COMPILE_TIME_ASSERT( ( 360 % NUM_SOLUTIONS ) == 0 );
struct Solution_t { // The sum bias
float bias; float highBias; AI_MoveSuggestion_t *pHighSuggestion; };
Solution_t solutions[NUM_SOLUTIONS] = { { 0, 0, NULL } };
//---------------------------------
// The first thing we do is reweight and normalize the weights into a range of [-1..1], where
// a negative weight is a repulsion. This becomes a bias for the solver.
// @TODO (toml 06-18-02): this can be made sligtly more optimal by precalculating regulation adjusted weights
Assert( suggestions.Count() >= 1 ); NormalizeSuggestions( &suggestions[0], (&suggestions[0]) + suggestions.Count() );
//
// Add the biased suggestions to the solutions
//
for ( int iSuggestion = 0; iSuggestion < suggestions.Count(); ++iSuggestion ) { AI_MoveSuggestion_t ¤t = suggestions[iSuggestion];
// Convert arc values to solution indices relative to right post. Right is angle down, left is angle up.
float halfSpan = current.arc.span * 0.5; int center = V_round( ( halfSpan * NUM_SOLUTIONS ) / 360 ); int left = ( current.arc.span * NUM_SOLUTIONS ) / 360;
float angRight = current.arc.center - halfSpan;
if (angRight < 0.0) angRight += 360;
int base = ( angRight * NUM_SOLUTIONS ) / 360;
// Sweep from left to right, summing the bias. For positive suggestions,
// the bias is further weighted to favor the center of the arc.
const float positiveDegradePer180 = 0.05; // i.e., lose 5% of weight by the time hit 180 degrees off center
const float positiveDegrade = ( positiveDegradePer180 / ( NUM_SOLUTIONS * 0.5 ) );
for ( int i = 0; i < left + 1; ++i ) { float bias = 0.0;
if ( current.weight > 0) { int iOffset = center - i; float degrade = abs( iOffset ) * positiveDegrade;
if ( ( (current.flags & AIMS_FAVOR_LEFT ) && i > center ) || ( (current.flags & AIMS_FAVOR_RIGHT) && i < center ) ) { degrade *= 0.9; }
bias = current.weight - ( current.weight * degrade ); } else bias = current.weight;
int iCurSolution = (base + i) % NUM_SOLUTIONS;
solutions[iCurSolution].bias += bias; if ( bias > solutions[iCurSolution].highBias ) { solutions[iCurSolution].highBias = bias; solutions[iCurSolution].pHighSuggestion = ¤t; } } }
//
// Find the best solution
//
int best = -1; float biasBest = 0;
for ( int i = 0; i < NUM_SOLUTIONS; ++i ) { if ( solutions[i].bias > biasBest ) { best = i; biasBest = solutions[i].bias; } }
if ( best == -1 ) return false; // no solution
//
// Construct the results
//
float result = best * SOLUTION_ANG;
// If the matching suggestion is within the solution, use that as the result,
// as it is valid and more precise.
const float suggestionCenter = solutions[best].pHighSuggestion->arc.center;
if ( suggestionCenter > result && suggestionCenter <= result + SOLUTION_ANG ) result = suggestionCenter;
pResult->dir = result;
return true; }
//-------------------------------------
// Purpose: Adjusts the suggestion weights according to the type of the suggestion,
// apply the appropriate sign, ensure values are in expected ranges
//-------------------------------------
struct AI_MoveSuggWeights { float min; float max; };
static AI_MoveSuggWeights g_AI_MoveSuggWeights[] = // @TODO (toml 06-18-02): these numbers need tuning
{ { 0.20, 1.00 }, // AIMST_MOVE
{ -0.00, -0.25 }, // AIMST_AVOID_DANGER
{ -0.00, -0.25 }, // AIMST_AVOID_OBJECT
{ -0.00, -0.25 }, // AIMST_AVOID_NPC
{ -0.00, -0.25 }, // AIMST_AVOID_WORLD
{ -1.00, -1.00 }, // AIMST_NO_KNOWLEDGE
{ -0.60, -0.60 }, // AIMST_OSCILLATION_DETERRANCE
{ 0.00, 0.00 }, // AIMST_INVALID
};
void CAI_MoveSolver::NormalizeSuggestions( AI_MoveSuggestion_t *pBegin, AI_MoveSuggestion_t *pEnd ) { while ( pBegin != pEnd ) { const float min = g_AI_MoveSuggWeights[pBegin->type].min; const float max = g_AI_MoveSuggWeights[pBegin->type].max;
Assert( pBegin->weight >= -AIMS_EPS && pBegin->weight <= 1.0 + AIMS_EPS );
if ( pBegin->weight < AIMS_EPS ) // zero normalizes to zero
pBegin->weight = 0.0; else pBegin->weight = ( ( max - min ) * pBegin->weight ) + min;
while (pBegin->arc.center < 0) pBegin->arc.center += 360;
while (pBegin->arc.center >= 360) pBegin->arc.center -= 360;
++pBegin; } }
//-------------------------------------
bool CAI_MoveSolver::HaveRegulationForObstacle( CBaseEntity *pEntity) { for ( int i = 0; i < m_Regulations.Count(); ++i ) { if ( m_Regulations[i].hObstacleEntity != NULL && pEntity == m_Regulations[i].hObstacleEntity.Get() ) { return true; } } return false; }
//-----------------------------------------------------------------------------
//
// Commands and tests
//
#ifdef DEBUG
CON_COMMAND(ai_test_move_solver, "Tests the AI move solver system") { #ifdef DEBUG
const float EPS = 0.001; #endif
DevMsg( "Beginning move solver tests...\n" );
CAI_MoveSolver solver; AI_MoveSolution_t solution; int i;
//
// Value in, no regulations, should yield value out
//
{ DevMsg( "Simple... " );
for (i = 0; i < 360; ++i) { Assert( solver.Solve( AI_MoveSuggestion_t( AIMST_MOVE, 1, i, 180 ), &solution ) ); Assert( solution.dir == (float)i );
}
DevMsg( "pass.\n" ); solver.ClearRegulations(); }
//
// Two values in, should yield the first
//
{ DevMsg( "Two positive... " );
AI_MoveSuggestion_t suggestions[2];
suggestions[0].Set( AIMST_MOVE, 1.0, 180, 100 ); suggestions[1].Set( AIMST_MOVE, 0.5, 0, 100 );
Assert( solver.Solve( suggestions, 2, &solution ) ); Assert( solution.dir == (float)suggestions[0].arc.center );
DevMsg( "pass.\n" ); solver.ClearRegulations(); }
//
// Two values in, first regulated, should yield the second
//
{ DevMsg( "Avoid one of two... " );
AI_MoveSuggestion_t suggestions[2];
solver.AddRegulation(AI_MoveSuggestion_t( AIMST_AVOID_OBJECT, 1, 260, 60 ) );
suggestions[0].Set( AIMST_MOVE, 1.0, 270, 45 ); suggestions[1].Set( AIMST_MOVE, 1.0, 0, 45 );
Assert( solver.Solve( suggestions, 2, &solution ) ); Assert( solution.dir == (float)suggestions[1].arc.center );
DevMsg( "pass.\n" ); solver.ClearRegulations(); }
//
// No solution
//
{ DevMsg( "No solution... " );
AI_MoveSuggestion_t suggestions[2];
suggestions[0].Set( AIMST_MOVE, 1.0, 270, 90 ); suggestions[1].Set( AIMST_AVOID_OBJECT, 1.0, 260, 180 );
Assert( !solver.Solve( suggestions, 2, &solution ) );
DevMsg( "pass.\n" ); solver.ClearRegulations(); }
//
// Nearest solution, in tolerance
//
{ DevMsg( "Nearest solution, in tolerance... " );
AI_MoveSuggestion_t suggestions[2];
suggestions[0].Set( AIMST_MOVE, 1.0, 278, 90 ); suggestions[1].Set( AIMST_AVOID_OBJECT, 1.0, 260, 24 );
Assert( solver.Solve( suggestions, 2, &solution ) ); Assert( solution.dir == (float)suggestions[0].arc.center );
DevMsg( "pass.\n" ); solver.ClearRegulations(); }
//
// Nearest solution
//
{ DevMsg( "Nearest solution... " );
AI_MoveSuggestion_t suggestions[2];
suggestions[0].Set( AIMST_MOVE, 1.0, 270, 90 ); suggestions[1].Set( AIMST_AVOID_OBJECT, 1.0, 260, 40 );
Assert( solver.Solve( suggestions, 2, &solution ) ); Assert( solution.dir - 282 < EPS ); // given 60 solutions
DevMsg( "pass.\n" ); solver.ClearRegulations(); }
} #endif
//=============================================================================
|