Merge branch 'master' into feat/demo_refactoring

README.md

@@ -97,8 +97,11 @@ wandb login
 
 #### OpenManipulator
 
+Follow the ROS installation commands in [Dockerfile](https://github.com/kairproject/kair_algorithms_draft/blob/master/Dockerfile) to train.
+
 ```
-python run_open_manipulator_reacher_v0.py --algo [algo] --off-render --log
+roslaunch kair_algorithms open_manipulator_env.launch gui:=false &
+rosrun run_open_manipulator_reacher_v0.py --algo [algo] --off-render --log
 ```
 
 #### LunarLanderContinuous-v2
@@ -113,6 +116,27 @@ python run_lunarlander_continuous.py --algo [algo] --off-render --log
 python run_reacher_v1.py --algo [algo] --off-render --log
 ```
 
+## How to Test
+
+#### OpenManipulator
+
+```
+roslaunch kair_algorithms open_manipulator_env.launch gui:=false &
+rosrun python run_open_manipulator_reacher_v0.py --algo [algo] --off-render --test --load-from [trained_weight_path]
+```
+
+#### LunarLanderContinuous-v2
+
+```
+python run_lunarlander_continuous.py --algo [algo] --off-render --test --load-from [trained_weight_path]
+```
+
+#### Reacher-v1
+
+```
+python run_reacher_v1.py --algo [algo] --off-render --test --load-from [trained_weight_path]
+```
+
 ## How to Cite
 
 We are currently writing a white paper to summarize the results. We will add a BibTeX entry below once the paper is finalized.

scripts/envs/open_manipulator/open_manipulator_reacher_env.py

@@ -76,15 +76,16 @@ class OpenManipulatorReacherEnv(gym.Env):
         if self.env_mode == "sim":
             self.reward = self.compute_reward()
             # TODO: Add termination condition
-            #            if self.ros_interface.check_for_termination():
-            #                self.done = True
+            # if self.ros_interface.check_for_termination():
+            #     self.done = True
             if self.ros_interface.check_for_success():
                 self.done = True
+                self.episode_steps = 0
 
         obs = self.ros_interface.get_observation()
 
         if self.episode_steps == self._max_episode_steps:
-            self.done = True
+            self.done = False
             self.episode_steps = 0
 
         return obs, self.reward_rescale_ratio * self.reward, self.done, None

scripts/envs/open_manipulator/ros_interface.py

@@ -306,7 +306,6 @@ class OpenManipulatorRosBaseInterface(object):
             self.success_count += 1
             if self.success_count == self.cfg["SUCCESS_COUNT"]:
                 print ("Current episode succeeded")
-                self.success_count = 0
                 return True
             else:
                 return False
@@ -388,6 +387,7 @@ class OpenManipulatorRosGazeboInterface(OpenManipulatorRosBaseInterface):
 
         # self.delete_target_block()
         self.init_robot_pose()
+        self.success_count = 0
         time.sleep(0.5)
 
         self.set_target_block(block_pose)