def get_plan(self):
     '''
     Calculate the plan from current image to the goal. 
     '''
     print('Received a request for plan to goal')
     
     
     if self.y_current is None or self.y_goal is None:
         print('Unable to plan because missing images.')
         return [-1]
     
     
     # Plan
     instance = namedtuple('instance', ['y_stary', 'y_goal'])
     
     size = self.size
     instance.y_start = get_uncertain_image(self.y_current, size)
     instance.y_goal = get_uncertain_image(self.y_goal, size)
     
     # Run planning
     result = self.algo.plan(instance.y_start, instance.y_goal)
     plan = result.plan
     
     print('Planning done, found plan: ' + str(plan))
     
     # Fix a bug temporarily TODO: Relearn diffeos
     orbit = OrbitModule(size)
     
     return orbit.inverse_plan(plan)
    def get_plan(self):
        '''
        Calculate the plan from current image to the goal. 
        '''
        print('Received a request for plan to goal')

        if self.y_current is None or self.y_goal is None:
            print('Unable to plan because missing images.')
            return [-1]

        # Plan
        instance = namedtuple('instance', ['y_stary', 'y_goal'])

        size = self.size
        instance.y_start = get_uncertain_image(self.y_current, size)
        instance.y_goal = get_uncertain_image(self.y_goal, size)

        # Run planning
        result = self.algo.plan(instance.y_start, instance.y_goal)
        plan = result.plan

        print('Planning done, found plan: ' + str(plan))

        # Fix a bug temporarily TODO: Relearn diffeos
        orbit = OrbitModule(size)

        return orbit.inverse_plan(plan)
Ejemplo n.º 3
0
    def __init__(self, id_discdds, diffeo_structure_threshold, id_algo,
                 plan_length, num_tests, get_planning_thresholds, plans):

        config = get_current_config()
        self.config = config

        # Load objects from configuration manager
        self.discdds = config.discdds.instance(id_discdds)
        self.algo = init_algorithm(self.config, id_algo, id_discdds,
                                   self.discdds)
        self.cmdlist = [
            self.discdds.actions[i].original_cmd
            for i in range(len(self.discdds.actions))
        ]
        #        pdb.set_trace()
        self.metric_goal = self.algo.metric_goal
        # Save input arguments
        self.id_discdds = id_discdds
        self.id_algo = id_algo
        self.diffeo_structure_threshold = diffeo_structure_threshold
        self.plan_length = plan_length
        self.num_tests = num_tests

        # Load orbit camera module
        self.orbit_module = OrbitModule(self.discdds.get_shape())

        # Set get_planning_thresholds function
        if get_planning_thresholds.__class__ == 'str':
            try:
                self.get_planning_thresholds = eval(get_planning_thresholds)
            except:
                logger.info()
        else:
            self.get_planning_thresholds = get_planning_thresholds
        self.get_planning_thresholds_name = str(get_planning_thresholds)

        # Initiate diffeo_structure
        self.diffeo_struct = DiffeoStructure(self.discdds,
                                             diffeo_structure_threshold)

        if plans == 'random':
            self.plans = self.gennerate_random_plans()
        elif plans.__class__ in [list, tuple]:
            self.plans = tuple(plans)
        else:
            self.plans = eval(plans)

        logger.info('Initialized with plans: ' + str(self.plans))
Ejemplo n.º 4
0
    def __init__(self, id_discdds,
                 diffeo_structure_threshold,
                 id_algo,
                 plan_length,
                 num_tests,
                 get_planning_thresholds,
                 plans):
        
        config = get_current_config()
        self.config = config
        
        
        # Load objects from configuration manager
        self.discdds = config.discdds.instance(id_discdds)
        self.algo = init_algorithm(self.config, id_algo, id_discdds, self.discdds)
        self.cmdlist = [self.discdds.actions[i].original_cmd for i in range(len(self.discdds.actions))]
#        pdb.set_trace()
        self.metric_goal = self.algo.metric_goal 
        # Save input arguments
        self.id_discdds = id_discdds
        self.id_algo = id_algo
        self.diffeo_structure_threshold = diffeo_structure_threshold
        self.plan_length = plan_length
        self.num_tests = num_tests
        
        # Load orbit camera module
        self.orbit_module = OrbitModule(self.discdds.get_shape())
        
        # Set get_planning_thresholds function
        if get_planning_thresholds.__class__ == 'str':
            try:
                self.get_planning_thresholds = eval(get_planning_thresholds)
            except:
                logger.info()  
        else:
            self.get_planning_thresholds = get_planning_thresholds
        self.get_planning_thresholds_name = str(get_planning_thresholds)
        
        # Initiate diffeo_structure
        self.diffeo_struct = DiffeoStructure(self.discdds, diffeo_structure_threshold)
        
        if plans == 'random':
            self.plans = self.gennerate_random_plans()
        elif plans.__class__ in [list, tuple]:
            self.plans = tuple(plans) 
        else:
            self.plans = eval(plans)
            
        logger.info('Initialized with plans: ' + str(self.plans))
Ejemplo n.º 5
0
class OnlinePlanning():
    def __init__(self, id_discdds,
                 diffeo_structure_threshold,
                 id_algo,
                 plan_length,
                 num_tests,
                 get_planning_thresholds,
                 plans):
        
        config = get_current_config()
        self.config = config
        
        
        # Load objects from configuration manager
        self.discdds = config.discdds.instance(id_discdds)
        self.algo = init_algorithm(self.config, id_algo, id_discdds, self.discdds)
        self.cmdlist = [self.discdds.actions[i].original_cmd for i in range(len(self.discdds.actions))]
#        pdb.set_trace()
        self.metric_goal = self.algo.metric_goal 
        # Save input arguments
        self.id_discdds = id_discdds
        self.id_algo = id_algo
        self.diffeo_structure_threshold = diffeo_structure_threshold
        self.plan_length = plan_length
        self.num_tests = num_tests
        
        # Load orbit camera module
        self.orbit_module = OrbitModule(self.discdds.get_shape())
        
        # Set get_planning_thresholds function
        if get_planning_thresholds.__class__ == 'str':
            try:
                self.get_planning_thresholds = eval(get_planning_thresholds)
            except:
                logger.info()  
        else:
            self.get_planning_thresholds = get_planning_thresholds
        self.get_planning_thresholds_name = str(get_planning_thresholds)
        
        # Initiate diffeo_structure
        self.diffeo_struct = DiffeoStructure(self.discdds, diffeo_structure_threshold)
        
        if plans == 'random':
            self.plans = self.gennerate_random_plans()
        elif plans.__class__ in [list, tuple]:
            self.plans = tuple(plans) 
        else:
            self.plans = eval(plans)
            
        logger.info('Initialized with plans: ' + str(self.plans))
        
    def gennerate_random_plans(self):
        
        n_cmd = len(self.discdds.actions)
        plans = []
        for _ in range(self.num_tests):
            plan = ()
            ie = 0
            while len(plan) < int(self.plan_length):
                # Add new command
                plan += (np.random.randint(0, n_cmd),)
                # remove redundant command
                plan = self.diffeo_struct.get_canonical(plan)
                ie += 1
                if ie > 50:
                    print('WARNING, a lot of iterations to generate a non redundant plan.')
                    plans.append(plan)
                    break
                
            plans.append(plan)
        return plans
       
    def run_all_tests(self, env='default'):
        all_stats = []
        for plan in self.plans:
            stat = self.run_test(plan, env)
            all_stats.append(stat)
        return all_stats

    def run_test(self, plan, env='default'):
        # Initiate stats object
        labels = {}
        labels['id_discdds'] = self.id_discdds
        labels['diffeo_structure_threshold'] = self.diffeo_structure_threshold
        labels['id_algo'] = self.id_algo
        labels['plan_length'] = self.plan_length
        labels['get_planning_thresholds'] = self.get_planning_thresholds_name
        labels['env'] = env
        active_instance = OnlineStats(labels, self.metric_goal)
        
        # Setup the planning problem
        self.create_testcase(plan_true=plan,
                             active_instance=active_instance)
        
        # Run the planning problem
        self.run_planning(active_instance=active_instance)
        
        self.prediction_images(active_instance)
        # return info about test
        return active_instance
    
    def create_testcase(self, plan_true, active_instance):
        # The goal image is where we start the demo
#        self.y_goal = self.orbit_module.get_image()
        active_instance.y_goal = self.orbit_module.get_image()
        active_instance.labels['plan_true'] = plan_true
        active_instance.labels['plan_true_reduced'] = self.diffeo_struct.get_canonical(plan_true)
        active_instance.plan_true = plan_true
        active_instance.plan_true_reduced = self.diffeo_struct.get_canonical(plan_true)
        
        # Move the camera to the start position
        plan_inverse = self.orbit_module.inverse_plan(plan_true)
        self.orbit_module.execute_plan(plan_inverse)
        
#        self.y_start =  self.orbit_module.get_image()
    


    
    def run_planning(self, active_instance):
        # Capture the initial image
        active_instance.y_start = self.orbit_module.get_image()
#        pdb.set_trace()
        if not self.get_planning_thresholds.__class__ in [list, tuple]:
            precision, min_visibility = self.get_planning_thresholds(algo=self.algo,
                                                                     discdds=self.discdds,
                                                                     active_instance=active_instance)
        else:
            precision, min_visibility = self.get_planning_thresholds
            
        active_instance.labels['precision'] = precision 
        active_instance.labels['min_visibility'] = min_visibility
        
        active_instance.precision = precision 
        active_instance.min_visibility = min_visibility
        
        planning_result = self.algo.plan(active_instance.y_start, active_instance.y_goal, precision=precision,
                                    min_visibility=min_visibility)
        if not planning_result.success:
            return None
        logger.info('Plan found: ' + str(planning_result.plan))

        plan_found = self.orbit_module.inverse_plan(planning_result.plan)
        self.orbit_module.execute_plan(plan_found)
        
        # Update active_instance with results
        active_instance.plan_found = plan_found
        active_instance.plan_found_reduced = self.diffeo_struct.get_canonical(plan_found)
        active_instance.y_result = self.orbit_module.get_image()
        
        active_instance.labels['plan_found'] = plan_found
        active_instance.labels['plan_found_reduced'] = self.diffeo_struct.get_canonical(plan_found)
        

    def prediction_images(self, active_instance):
        '''
        predicts the resulting images from active_instance.y_start and plan 
        :param active_instance:
        '''
        if active_instance.plan_found is not None:
            inv_plan = self.orbit_module.inverse_plan(active_instance.plan_found)
            active_instance.y_found_pred = self.discdds.predict(active_instance.y_start, inv_plan)
            
        if active_instance.plan_true is not None:
            inv_plan = self.orbit_module.inverse_plan(active_instance.plan_true)
            active_instance.y_goal_pred = self.discdds.predict(active_instance.y_start, inv_plan)
Ejemplo n.º 6
0
class OnlinePlanning():
    def __init__(self, id_discdds, diffeo_structure_threshold, id_algo,
                 plan_length, num_tests, get_planning_thresholds, plans):

        config = get_current_config()
        self.config = config

        # Load objects from configuration manager
        self.discdds = config.discdds.instance(id_discdds)
        self.algo = init_algorithm(self.config, id_algo, id_discdds,
                                   self.discdds)
        self.cmdlist = [
            self.discdds.actions[i].original_cmd
            for i in range(len(self.discdds.actions))
        ]
        #        pdb.set_trace()
        self.metric_goal = self.algo.metric_goal
        # Save input arguments
        self.id_discdds = id_discdds
        self.id_algo = id_algo
        self.diffeo_structure_threshold = diffeo_structure_threshold
        self.plan_length = plan_length
        self.num_tests = num_tests

        # Load orbit camera module
        self.orbit_module = OrbitModule(self.discdds.get_shape())

        # Set get_planning_thresholds function
        if get_planning_thresholds.__class__ == 'str':
            try:
                self.get_planning_thresholds = eval(get_planning_thresholds)
            except:
                logger.info()
        else:
            self.get_planning_thresholds = get_planning_thresholds
        self.get_planning_thresholds_name = str(get_planning_thresholds)

        # Initiate diffeo_structure
        self.diffeo_struct = DiffeoStructure(self.discdds,
                                             diffeo_structure_threshold)

        if plans == 'random':
            self.plans = self.gennerate_random_plans()
        elif plans.__class__ in [list, tuple]:
            self.plans = tuple(plans)
        else:
            self.plans = eval(plans)

        logger.info('Initialized with plans: ' + str(self.plans))

    def gennerate_random_plans(self):

        n_cmd = len(self.discdds.actions)
        plans = []
        for _ in range(self.num_tests):
            plan = ()
            ie = 0
            while len(plan) < int(self.plan_length):
                # Add new command
                plan += (np.random.randint(0, n_cmd), )
                # remove redundant command
                plan = self.diffeo_struct.get_canonical(plan)
                ie += 1
                if ie > 50:
                    print(
                        'WARNING, a lot of iterations to generate a non redundant plan.'
                    )
                    plans.append(plan)
                    break

            plans.append(plan)
        return plans

    def run_all_tests(self, env='default'):
        all_stats = []
        for plan in self.plans:
            stat = self.run_test(plan, env)
            all_stats.append(stat)
        return all_stats

    def run_test(self, plan, env='default'):
        # Initiate stats object
        labels = {}
        labels['id_discdds'] = self.id_discdds
        labels['diffeo_structure_threshold'] = self.diffeo_structure_threshold
        labels['id_algo'] = self.id_algo
        labels['plan_length'] = self.plan_length
        labels['get_planning_thresholds'] = self.get_planning_thresholds_name
        labels['env'] = env
        active_instance = OnlineStats(labels, self.metric_goal)

        # Setup the planning problem
        self.create_testcase(plan_true=plan, active_instance=active_instance)

        # Run the planning problem
        self.run_planning(active_instance=active_instance)

        self.prediction_images(active_instance)
        # return info about test
        return active_instance

    def create_testcase(self, plan_true, active_instance):
        # The goal image is where we start the demo
        #        self.y_goal = self.orbit_module.get_image()
        active_instance.y_goal = self.orbit_module.get_image()
        active_instance.labels['plan_true'] = plan_true
        active_instance.labels[
            'plan_true_reduced'] = self.diffeo_struct.get_canonical(plan_true)
        active_instance.plan_true = plan_true
        active_instance.plan_true_reduced = self.diffeo_struct.get_canonical(
            plan_true)

        # Move the camera to the start position
        plan_inverse = self.orbit_module.inverse_plan(plan_true)
        self.orbit_module.execute_plan(plan_inverse)


#        self.y_start =  self.orbit_module.get_image()

    def run_planning(self, active_instance):
        # Capture the initial image
        active_instance.y_start = self.orbit_module.get_image()
        #        pdb.set_trace()
        if not self.get_planning_thresholds.__class__ in [list, tuple]:
            precision, min_visibility = self.get_planning_thresholds(
                algo=self.algo,
                discdds=self.discdds,
                active_instance=active_instance)
        else:
            precision, min_visibility = self.get_planning_thresholds

        active_instance.labels['precision'] = precision
        active_instance.labels['min_visibility'] = min_visibility

        active_instance.precision = precision
        active_instance.min_visibility = min_visibility

        planning_result = self.algo.plan(active_instance.y_start,
                                         active_instance.y_goal,
                                         precision=precision,
                                         min_visibility=min_visibility)
        if not planning_result.success:
            return None
        logger.info('Plan found: ' + str(planning_result.plan))

        plan_found = self.orbit_module.inverse_plan(planning_result.plan)
        self.orbit_module.execute_plan(plan_found)

        # Update active_instance with results
        active_instance.plan_found = plan_found
        active_instance.plan_found_reduced = self.diffeo_struct.get_canonical(
            plan_found)
        active_instance.y_result = self.orbit_module.get_image()

        active_instance.labels['plan_found'] = plan_found
        active_instance.labels[
            'plan_found_reduced'] = self.diffeo_struct.get_canonical(
                plan_found)

    def prediction_images(self, active_instance):
        '''
        predicts the resulting images from active_instance.y_start and plan 
        :param active_instance:
        '''
        if active_instance.plan_found is not None:
            inv_plan = self.orbit_module.inverse_plan(
                active_instance.plan_found)
            active_instance.y_found_pred = self.discdds.predict(
                active_instance.y_start, inv_plan)

        if active_instance.plan_true is not None:
            inv_plan = self.orbit_module.inverse_plan(
                active_instance.plan_true)
            active_instance.y_goal_pred = self.discdds.predict(
                active_instance.y_start, inv_plan)
Ejemplo n.º 7
0
 def get_plan(self, req):
     '''
     Calculate the plan from current image to the goal.
     Called from ROS srv. 
     '''
     self.info(str(req))
     self.info('Received a request for plan to goal')
     
     
     # Initiate response object
     res = camera_actuator.srv.planServiceResponse()  # @UndefinedVariable
     
     if self.state in [WAIT_GOAL, WAIT_CURRENT]:
         self.info('Module is in state: ' + self.state + 
                   'Unable to plan because of unspecified images.')
         res.plan = [-1]
         return res
     
     # Assert the module is idle state, otherwise refuse to plan.
     if not self.state == IDLE:
         self.info('Module is in state: ' + self.state + 
                   ' and will not accept new requests for plans right now')
         res.plan = [-1]
         return res
     
     self.info('Passed state check, OK')
     self.state = PLANNING  # indicate that the module is busy with planning
     
     # Plan
     instance = namedtuple('instance', ['y_stary', 'y_goal'])
     
     size = self.size
     y_start = self.get_y_current(size)
     y_goal = self.get_y_goal(size)
     instance.y_start = y_start
     instance.y_goal = y_goal
     
     # Run planning
     algo = self.online_planner.algo
     discdds = self.online_planner.discdds
     
     thresholds = rospy.get_param(self.node_name + '/planning_thresholds')
     if thresholds.__class__ == str:
         get_planning_thresholds = eval('online_planning.' + thresholds)
         thresh = get_planning_thresholds(algo=algo,
                                          discdds=discdds,
                                          active_instance=instance)
     else:
         thresh = thresholds
         
     precision, min_visibility = thresh
     self.info('Ready to plan with precision %g' % precision)
     
     result = self.online_planner.algo.plan(y_start,
                                            y_goal,
                                            precision,
                                            min_visibility)
     plan = result.plan
     
     self.info('Planning done, found plan: ' + str(plan))
     
     orbit = OrbitModule(size)
     
     
     res.plan = orbit.inverse_plan(plan)
     self.state = IDLE
     return res