Package Metadata & CV Model Update

- Remove commented out code from rrrobot.yaml and change time_limit to 10 minutes
- Update dependencies in rrrobot package.xml
- Update ROS package version to 1.0.0
- Update CV model with better pre-trained model

src/rrrobot_ws/src/rrrobot/config/rrrobot.yaml

@@ -10,154 +10,5 @@ options:
     order_part3: pulley_part
   visualize_drop_regions: false     # Whether or not to visualize drop regions (world frame only)
 
-time_limit: 500                      # Maximum time allowed for the trial once started, in seconds
+time_limit: 600                      # Maximum time allowed for the trial once started, in seconds
 random_seed: 1                       # Seed for the pseudo random number generator (used to randomize model names)
-
-# Orders to announce during the trial
-# orders:
-#   order_0:
-#     announcement_condition: time     # Announce the order base on elapsed time
-#     announcement_condition_value: 0  # Time in seconds to wait before announcing the order
-#     shipment_count: 1                # How many of the same shipment are in the order
-#     destinations: [agv1]             # Which agv the shipments must be delivered to
-#                                      #  If specified it must be a list the same length as the shipment count
-#                                      #  allowed values: agv1, agv2, any
-#     products:                        # List of products required to be in the kit
-#       part_0:
-#         type: order_part1            # Type of model required
-#         pose:
-#           xyz: [0.1, -0.15, 0]       # Position required in the tray frame
-#           rpy: [0, 0, 'pi/2']        # Orientation required in the tray frame
-#       part_1:
-#         type: order_part1
-#         pose:
-#           xyz: [-0.1, -0.15, 0]
-#           rpy: [0, 0, 'pi/2']
-#       part_2:
-#         type: order_part2
-#         pose:
-#           xyz: [0.1, 0.15, 0]
-#           rpy: [0, 0, 0]
-#       part_3:
-#         type: order_part2
-#         pose:
-#           xyz: [-0.1, 0.15, 0]
-#           rpy: [0, 0, 0]
-#   order_1:
-#     announcement_condition: wanted_products # Announce the order when the boxes contain products from this order
-#     announcement_condition_value: 2
-#     shipment_count: 1
-#     destinations: [any]
-#     products:
-#       part_0:
-#         type: order_part3
-#         pose:
-#           xyz: [0.12, -0.2, 0]
-#           rpy: ['pi', 0, 0]
-#       part_1:
-#         type: order_part3
-#         pose:
-#           xyz: [-0.12, -0.2, 0]
-#           rpy: [0, 'pi', 0]
-#       part_2:
-#         type: order_part1
-#         pose:
-#           xyz: [0.15, 0.15, 0]
-#           rpy: [0, 0, 0]
-#       part_3:
-#         type: order_part2
-#         pose:
-#           xyz: [-0.15, 0.15, 0]
-#           rpy: [0, 0, 0]
-#       part_4:
-#         type: order_part2
-#         pose:
-#           rpy: [0, 'pi', 0]
-
-# # Individual products that will be reported as faulty
-# faulty_products:
-#   - piston_rod_part_57                # The piston rod part in the bins with randomized ID of 57
-#   - piston_rod_part_45
-
-# # Models to be inserted in the bins
-# models_over_bins:
-#   bin1:                              # Name of the bin (bin1-bin6, as named in the environment simulation)
-#     models:                          # List of models to insert
-#       gear_part:                     # Type of model to insert
-#         xyz_start: [0.1, 0.1, 0]     # Origin of the first model to insert
-#         xyz_end: [0.5, 0.5, 0]       # Origin of the last model to insert
-#         rpy: [0, 0, 'pi/4']          # Orientation of all models to insert
-#         num_models_x: 3              # How many models to insert along the x dimension
-#         num_models_y: 5              # How many models to insert along the y dimension
-#   bin5:
-#     models:
-#       gasket_part:
-#         xyz_start: [0.1, 0.1, 0.0]
-#         xyz_end: [0.5, 0.5, 0.0]
-#         rpy: [0, 0, 'pi/4']
-#         num_models_x: 2
-#         num_models_y: 3
-#   bin6:
-#     models:
-#       piston_rod_part:
-#         xyz_start: [0.1, 0.1, 0.0]
-#         xyz_end: [0.5, 0.5, 0.0]
-#         rpy: [0, 0, 'pi/4']
-#         num_models_x: 3
-#         num_models_y: 3
-
-# # Models to be spawned in particular reference frames
-# models_to_spawn:
-#   bin2::link:                        # Name of the reference frame
-#     models:                          # List of models to spawn
-#       piston_rod_part_1:             # An arbitrary unique name of the model (will be randomized)
-#         type: piston_rod_part        # Type of model (must be installed locally)
-#         pose:
-#           xyz: [0.2, 0.2, 0.75]      # Co-ordinates of the model origin in the specified reference frame
-#           rpy: [0, 0, '-pi/2']       # Roll, pitch, yaw of the model in the specified reference frame
-
-# # Models to be spawned on the conveyor belt
-# belt_models:
-#   pulley_part:                   # name of part model to spawn
-#     1.0:                             # Time in seconds after trial starts to spawn it
-#       pose:
-#         xyz: [0.0, 0.0, 0.1]         # Coordinates relative to the start of the container to spawn the part
-#         rpy: [0, 0, 'pi/2']          # Roll, pitch, yaw of the model
-#   disk_part:
-#     5.0:
-#       pose:
-#         xyz: [0.0, 0.0, 0.1]
-#         rpy: [0, 0, 'pi/2']
-
-# # Drops from the vacuum gripper to be triggered at particular locations
-# drops:
-#   drop_regions:
-#     above_agv_1:
-#       frame: agv1::kit_tray_1        # Frame the drop region/destination is specified in
-#       min:                           # Min corner of the bounding box that triggers a drop
-#         xyz: [-0.3, -0.3, 0.0]
-#       max:                           # Max corner of the bounding box that triggers a drop
-#         xyz: [0.3, 03, 0.5]
-#       destination:                   # Where to drop the part to
-#         xyz: [-0.35, 0.1, 0.15]
-#         rpy: [0, 0, 0.2]
-#       product_type_to_drop: gear_part
-#     above_agv_2:
-#       frame: agv2::kit_tray_2
-#       min:
-#         xyz: [-0.3, -0.3, 0.0]
-#       max:
-#         xyz: [0.3, 0.3, 0.5]
-#       destination:
-#         xyz: [0.15, 0.15, 0.15]
-#         rpy: [0, 0, -0.2]
-#       product_type_to_drop: pulley_part
-#     bin5_reachable:
-#       min:
-#         xyz: [0.0, 0.7, 0.7]
-#       max:
-#         xyz: [0.2, 1.60, 1.3]
-#       destination:
-#         xyz: [0.65, 1.15, 0.76]
-#         rpy: [0, 0, 0.5]
-#       product_type_to_drop: gasket_part

src/rrrobot_ws/src/rrrobot/package.xml

@@ -1,7 +1,7 @@
 <?xml version="1.0"?>
 <package format="2">
     <name>rrrobot</name>
-    <version>0.1.0</version>
+    <version>1.0.0</version>
     <description>RRRobot package</description>
     <maintainer email="balajsra@umich.edu">Sravan Balaji</maintainer>
     <maintainer email="chenxgu@umich.edu">Chenxi Gu</maintainer>
@@ -16,12 +16,17 @@
     <depend>roscpp</depend>
     <depend>sensor_msgs</depend>
     <depend>std_srvs</depend>
-    <depend>tf2_geometry_msgs</depend>
     <depend>trajectory_msgs</depend>
     <depend>std_msgs</depend>
     <depend>geometry_msgs</depend>
     <depend>gazebo_ros</depend>
     <depend>message_generation</depend>
+    <depend>pcl_ros</depend>
+    <depend>pcl_conversion</depend>
+    <depend>gazebo</depend>
+    <depend>PCL</depend>
+    <depend>Eigen3</depend>
+    <depend>orocos_kdl</depend>
     <exec_depend>message_runtime</exec_depend>
 
 </package>

src/rrrobot_ws/src/rrrobot/src/cv_model.py

@@ -8,19 +8,17 @@ from torch import nn
 from PIL import Image
 
 
-class Model_v1(nn.Module):
+class Model_mobilenet(nn.Module):
     def __init__(self):
-        super(Model_v1, self).__init__()
-        self.resnet = torchvision.models.resnet34(pretrained=True)
-        self.resnet.fc = torch.nn.Linear(512, 512)
-        self.addition = nn.Sequential(
-            nn.BatchNorm1d(512),
-            nn.Dropout(p=0.5),
-            nn.Linear(512, 6))
+        super(Model_mobilenet, self).__init__()
+        self.mobilenet = torchvision.models.mobilenet_v2(pretrained=True)
+
+        self.mobilenet.classifier = nn.Sequential(
+            nn.Dropout(p=0.2, inplace=False),
+            nn.Linear(1280, 6, bias=True))
 
     def forward(self, image):
-        z = self.resnet(image)
-        z = self.addition(z)
+        z = self.mobilenet(image)
         return z
 
 
@@ -28,6 +26,8 @@ def call_back(filename):
     # prepare input file
     filename = filename.data
     input_image = Image.open(filename)
+    if input_image.size[0] < input_image.size[1]:
+            input_image = input_image.transpose(Image.ROTATE_90)
     input_image = input_image.resize((512, 384))
     # print(input_image.size)
     input_image = np.moveaxis(np.asarray(input_image), 2, 0).astype(np.float32)
@@ -57,7 +57,7 @@ def call_back(filename):
     garbage_type = type_dict[predicted_label]
     rospy.loginfo(filename + ':' + garbage_type)
     pub.publish(filename + ':' + garbage_type)
-    rate.sleep()
+    # rate.sleep()
 
 
 def listener():
@@ -69,10 +69,10 @@ def listener():
 
 if __name__ == '__main__':
     rospy.init_node('cv_node')
-    
+
     # prepare pretrained model
-    model = Model_v1()
-    model.load_state_dict(torch.load('v1_0.001.pt', map_location='cpu'))
+    model = Model_mobilenet()
+    model.load_state_dict(torch.load('pytorch_pretrain_model.pt', map_location='cpu'))
     model = model.to('cpu')
     model.eval()