float WalkOptimisationProvider::calculateFitness()
{
#if defined(USE_SPEED)
return calculateFitnesses()[0];
#elif defined(USE_COST)
return calculateFitnesses()[1];
#endif
}
void WalkOptimisationProvider::tickOptimiser()
{
#if DEBUG_BEHAVIOUR_VERBOSITY > 1
debug << "WalkOptimisationProvider::tickOptimiser" << endl;
#endif
// update the optimiser and give the next set of parameters to the walk engine
#ifdef USE_MO
vector<float> fitness = calculateFitnesses();
#else
float fitness = calculateFitness();
#endif
if (m_optimiser)
{
m_optimiser->setParametersResult(fitness);
vector<float> nextparameters = m_optimiser->getNextParameters();
m_parameters.set(nextparameters);
}
m_jobs->addMotionJob(new WalkParametersJob(m_parameters));
#if DEBUG_BEHAVIOUR_VERBOSITY > 1
debug << "WalkOptimisationProvider::tickOptimiser() new parameters: " << m_parameters.getAsVector() << endl;
#endif
// save the state of the optimiser, and the walk parameters in case of hardware failure.
if (m_optimiser)
m_optimiser->save();
m_iteration_count++;
}
/**
* Selects the new generation by mating two individuals selected
* probabilistically by their fitness.
* Point 3 of page 7 in paper by Mitchell.
*/
void make_new_generation() {
int i,j,k;
int totalFitness = calculateFitnesses();
rule_t* next_population = populations[1-p_buf];
if (generation % GENERATION_PRINT_SEQUENCE == 0) {
printGeneration();
}
// Create new generation
for (i=0; i<POPULATION_SIZE; i++) {
int index1 = 0, index2 = 0;
int map1 = (int) random_max(totalFitness);
int map2 = (int) random_max(totalFitness);
bool done1 = false;
bool done2 = false;
// Find individuals to mate
for (j=0, k=0; j < POPULATION_SIZE; j++) {
if (done1 && done2) break;
k += population[j].fitness;
if (!done1 && k >= map1) {
index1 = j;
done1 = true;
}
if (!done2 && k >= map2) {
index2 = j;
done2 = true;
}
}
rule_mate(population+index1, population+index2, next_population+i, next_population+i+1);
i++;
}
p_buf = 1-p_buf;
population = next_population;
}
