changes

mushr_rhc_ros/src/rhcnode_network_shuang.py

@@ -78,8 +78,13 @@ def __init__(self, dtype, params, logger, name):
         device = torch.device('cuda')
 
         self.clip_len = 16
+
+        # fine-tuned in real life
+        saved_model_path = '/home/robot/weight_files/epoch0.pth.tar'
         # saved_model_path = '/home/rb/downloaded_models/epoch30.pth.tar'
-        saved_model_path = '/home/robot/weight_files/epoch15.pth.tar'
+
+        # best so far
+        # saved_model_path = '/home/robot/weight_files/epoch15.pth.tar'
         # saved_model_path = '/home/rb/hackathon_data/aml_outputs/log_output/gpt_resnet18_0/GPTgpt_resnet18_4gpu_2022-01-24_1642987604.6403077_2022-01-24_1642987604.640322/model/epoch15.pth.tar'
         # saved_model_path = '/home/rb/hackathon_data/aml_outputs/log_output/gpt_resnet18_8_exp2/GPTgpt_resnet18_8gpu_exp2_2022-01-25_1643076745.003202_2022-01-25_1643076745.0032148/model/epoch12.pth.tar'
         vocab_size = 100
@@ -161,7 +166,7 @@ def start(self):
                 self.last_action = self.apply_network()
                 self.q_actions.get()  # remove the oldest action from the queue
                 self.q_actions.put(self.last_action)
-                rospy.loginfo("Applied network: "+str(self.last_action))
+                # rospy.loginfo("Applied network: "+str(self.last_action))
                 self.compute_network = False
 
             self.publish_traj(self.default_speed, self.last_action)
@@ -218,7 +223,7 @@ def apply_network(self):
         t = t.to(self.device)
 
         finish_processing = time.time()
-        rospy.loginfo("processing delay: "+str(finish_processing-start))
+        # rospy.loginfo("processing delay: "+str(finish_processing-start))
 
         # organize the action input
         with torch.set_grad_enabled(False):