def computeManipulabilityCost(self): ''' Computes the manipulability cost on one step of the trajectory Input: -cost: cost at time t, float Output: -cost: cost at time t+1, float ''' dotq, q = getDotQAndQFromStateVector(self.arm.getState()) manip = self.arm.directionalManipulability(q,self.cartTarget) return 1-manip
def runTrajectory(self, x, y, foldername): ''' Generates trajectory from the initial position (x, y) Inputs: -x: abscissa of the initial position, float -y: ordinate of the initial position, float Output: -cost: the cost of the trajectory, float ''' #computes the articular position q1, q2 from the initial coordinates (x, y) q1, q2 = self.arm.mgi(x, y) #creates the state vector [dotq1, dotq2, q1, q2] q = createVector(q1,q2) state = np.array([0., 0., q1, q2]) #print("start state --------------: ",state) #computes the coordinates of the hand and the elbow from the position vector coordElbow, coordHand = self.arm.mgdFull(q) #assert(coordHand[0]==x and coordHand[1]==y), "Erreur de MGD" does not work because of rounding effects #initializes parameters for the trajectory i, t, cost = 0, 0, 0 self.stateEstimator.initStore(state) self.arm.setState(state) estimState = state dataStore = [] qtarget1, qtarget2 = self.arm.mgi(self.rs.XTarget, self.rs.YTarget) vectarget = [0.0, 0.0, qtarget1, qtarget2] #loop to generate next position until the target is reached while coordHand[1] < self.rs.YTarget and i < self.rs.numMaxIter: stepStore = [] #computation of the next muscular activation vector using the controller theta #print ("state :",self.arm.getState()) U = self.controller.computeOutput(estimState) if det: Unoisy = muscleFilter(U) else: Unoisy = getNoisyCommand(U,self.arm.musclesP.knoiseU) Unoisy = muscleFilter(Unoisy) #computation of the arm state realNextState = self.arm.computeNextState(Unoisy, self.arm.getState()) #computation of the approximated state tmpState = self.arm.getState() if det: estimNextState = realNextState else: U = muscleFilter(U) estimNextState = self.stateEstimator.getEstimState(tmpState,U) #print estimNextState self.arm.setState(realNextState) #computation of the cost cost += self.computeStateTransitionCost(Unoisy) ''' print "U =", U print "Unoisy =", Unoisy print "estimstate =", estimState #print "theta =", self.controller.theta if math.isnan(cost): print "NAN : U =", U print "NAN : Unoisy =", Unoisy print "NAN : estimstate =", estimState #print "NAN : theta =", self.controller.theta sys.exit() ''' #get dotq and q from the state vector dotq, q = getDotQAndQFromStateVector(tmpState) coordElbow, coordHand = self.arm.mgdFull(q) #print ("dotq :",dotq) #computation of the coordinates to check if the target is reach or not #code to save data of the trajectory #Note : these structures might be much improved if self.saveTraj == True: stepStore.append(vectarget) stepStore.append(estimState) stepStore.append(tmpState) stepStore.append(Unoisy) stepStore.append(np.array(U)) stepStore.append(estimNextState) stepStore.append(realNextState) stepStore.append([coordElbow[0], coordElbow[1]]) stepStore.append([coordHand[0], coordHand[1]]) #print ("before",stepStore) tmpstore = np.array(stepStore).flatten() row = [item for sub in tmpstore for item in sub] #print ("store",row) dataStore.append(row) estimState = estimNextState i += 1 t += self.rs.dt cost += self.computeFinalReward(t,coordHand) if self.saveTraj == True: filename = findFilename(foldername+"Log/","traj",".log") np.savetxt(filename,dataStore) ''' if coordHand[0] >= -self.sizeOfTarget/2 and coordHand[0] <= self.sizeOfTarget/2 and coordHand[1] >= self.rs.YTarget and i<230: foldername = pathDataFolder + "TrajRepository/" name = findFilename(foldername,"Traj",".traj") np.savetxt(name,dataStore) ''' lastX = -1000 #used to ignore dispersion when the target line is not crossed if coordHand[1] >= self.rs.YTarget: lastX = coordHand[0] #print "end of trajectory" return cost, t, lastX
def computePerpendCost(self): dotq, q = getDotQAndQFromStateVector(self.arm.getState()) J = self.arm.jacobian(q) xi = np.dot(J,dotq) xi = xi/np.linalg.norm(xi) return 500-1000*xi[0]*xi[0]