are you using roslaunch to run that launch file? (that seems like a python error)

from cadrl_ros.

are you using roslaunch to run that launch file? (that seems like a python error)

yes, I use 'roslaunch cadrl_node.launch' and the result is as follows:
`... logging to /home/hb/.ros/log/a3201dbc-4d69-11eb-a154-000c291ba878/roslaunch-ubuntu-20219.log
Checking log directory for disk usage. This may take awhile.
Press Ctrl-C to interrupt
Done checking log file disk usage. Usage is <1GB.
started roslaunch server http://ubuntu:42785/
SUMMARY

PARAMETERS

• /JA01/cadrl_node/jackal_speed: 1.2
• /rosdistro: kinetic
• /rosversion: 1.12.17
  NODES
  /JA01/
  cadrl_node (cadrl_ros/cadrl_node.py)
  ROS_MASTER_URI=http://localhost:11311
  process[JA01/cadrl_node-1]: started with pid [20236]
  hello world from cadrl_node.py
  2021-01-03 03:50:40.976400: I tensorflow/core/platform/cpu_feature_guard.cc:137] Your CPU supports instructions that this TensorFlow binary was not compiled to use: SSE4.1 SSE4.2 AVX AVX2 FMA
  ('\n*******\n*\nJackal speed:', 1.2, '\n**********\n******')
  Exception in thread Thread-5:
  Traceback (most recent call last):
  File "/usr/lib/python2.7/threading.py", line 801, in __bootstrap_inner
  self.run()
  File "/opt/ros/kinetic/lib/python2.7/dist-packages/rospy/timer.py", line 235, in run
  self._callback(TimerEvent(last_expected, last_real, current_expected, current_real, last_duration))
  File "/home/hb/catkin_ws/src/cadrl_ros/scripts/cadrl_node.py", line 389, in cbComputeActionGA3C
  predictions = self.nn.predict_p(obs, None)[0]
  TypeError: predict_p() takes exactly 2 arguments (3 given)`
  I don't know why and I don't change anything in the code.

from cadrl_ros.

ah i think I changed the API for the neural network at some point and forgot to update the ros node. That line should work if you get rid of the 2nd argument to predict_p?

from cadrl_ros.

ah i think I changed the API for the neural network at some point and forgot to update the ros node. That line should work if you get rid of the 2nd argument to predict_p?

yes, I have tried to get rid of the 2nd argument to predict_p and it doesn't have the previous bug but it does nothing. So I removed annotation ‘print "predictions", predictions’, then it loops infinitely and I don't know why it can loop and how to shutdown?

from cadrl_ros.

ah i think I changed the API for the neural network at some point and forgot to update the ros node. That line should work if you get rid of the 2nd argument to predict_p?

emmm, I click ‘ctrl+c’ and it shuts down and the result is as follows.

^C[JA01/cadrl_node-1] killing on exit
[INFO] [1609811953.441362]: [/JA01/cadrl_node] Shutting down.
[INFO] [1609811953.442130]: Stopped JA01's velocity.
predictions: [0.087332 0.05659726 0.00574056 0.00810037 0.02762621 0.02491285
0.12819767 0.11566862 0.06534816 0.21978024 0.26069608]
best action index: 10
shutting down processing monitor...
... shutting down processing monitor complete
done

Is this right? It need to stop manually instead of automatically by itself?

from cadrl_ros.

when you say it loops infinitely, is it publishing Twist commands? the ros node should run indefinitely and keep querying the NN at a fixed frequency

from cadrl_ros.

when you say it loops infinitely, is it publishing Twist commands? the ros node should run indefinitely and keep querying the NN at a fixed frequency

I don't think it is publishing Twist commands because the print results are the same
'predictions: [0.087332 0.05659726 0.00574056 0.00810037 0.02762621 0.02491285 0.12819767 0.11566862 0.06534816 0.21978024 0.26069608]
best action index: 10
'
and you say the ros node should run indefinitely and keep querying the NN at a fixed frequency, is that means I do right to shutdown manually by clicking ‘ctrl+c’ instead of automatically by itself？

from cadrl_ros.

from cadrl_ros.

what do you see when you rostopic echo the nn_cmd_vel topic? sure, you can shut it down whenever you want by clicking ctrl+c, but i would expect the vehicle command will stay the same as long as the input observation is the same. so something needs to change on the input topics to see a change in the selected action. normally i would start the ros node, let it run while the robot moves around, then at the end of the experiment ctrl+c the node.
…
On Mon, Jan 4, 2021 at 9:24 PM HanBing0802 @.***> wrote: when you say it loops infinitely, is it publishing Twist commands? the ros node should run indefinitely and keep querying the NN at a fixed frequency I don't think it is publishing Twist commands because the print results are the same 'predictions: [0.087332 0.05659726 0.00574056 0.00810037 0.02762621 0.02491285 0.12819767 0.11566862 0.06534816 0.21978024 0.26069608] best action index: 10 ' and you say the ros node should run indefinitely and keep querying the NN at a fixed frequency, is that means I do right to shutdown manually by clicking ‘ctrl+c’ instead of automatically by itself？ — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#13 (comment)>, or unsubscribe https://github.com/notifications/unsubscribe-auth/AA22BJJVJZ3TBEBJORGPYP3SYJZ4NANCNFSM4VRRCJEQ .
-- Michael Everett Postdoctoral Associate Department of Aeronautics and Astronautics Massachusetts Institute of Technology (MIT) mfe.mit.edu

when I run 'rostopic echo /nn_cmd_vel', it has a bug as follows:
WARNING: topic [/nn_cmd_vel] does not appear to be published yet

from cadrl_ros.

i think the topics are put under the /JA01 namespace by default in that launch file. so it would probably be rostopic echo /JA01/nn_cmd_vel -- but you can do rostopic list to see what topics are being published

from cadrl_ros.

i think the topics are put under the /JA01 namespace by default in that launch file. so it would probably be rostopic echo /JA01/nn_cmd_vel -- but you can do rostopic list to see what topics are being published

I am a beginner, so there are many problems, and thank you for your patience,
Yes, I try rostopic echo /JA01/nn_cmd_vel but it also loops infinitely and prints the same results
"linear:
x: 0.0
y: 0.0
z: 0.0
angular:
x: 0.0
y: 0.0
z: 0.0"
I think that means the agent doesn't move anymore. And then I run 'rosrun rqt_graph rqt_graph' and there are only a node '/JA01/cadrl_node' without any activate topics. So it's the final results? But I indeed don't know why they loop infinitely not only 'roslaunch cadrl_node.launch' but also 'rostopic echo /JA01/nn_cmd_vel'?

from cadrl_ros.

That seems right to me initially. The command should be to stay in place while waiting for a goal. Have you set a goal (eg by publishing a msg to the topic corresponding to the goal)? Once it receives a goal, the agent should spin in place (nonzero angular z) until aligning with the goal, then will move forward according to the NN output.

I think rqt_graph only shows topics that something is subscribing to

It loops infinitely because this ROS node is meant for a robot that continually receives new sensor data, and thus observations of agent states. If you're just looking for the policy on its own (eg for a single observation timestep, query the NN), there's a separate repo that would be simpler to use and not require ROS.

from cadrl_ros.

Hey Michael,
You means that if I just want to know the path of the robot to avoid obstacles, I need to publish a msg to set a goal position. Is that true? If so, I can understand why the output of rostopic echo /JA01/nn_cmd_vel is zero. Before this I always thought that the robot's path could be visualized after roslaunch cadrl_node.launch because I saw the function 'visualize_**()'. I'm sorry, because I didn't study ROS before, I took it for granted. I think next I should study ROS such as how to publish a msg.
Thank you very much!

from cadrl_ros.

That is correct -- the cadrl node needs to know the goal before it can compute velocities that will get the robot to that goal. I'll close this issue for now but feel free to re-open if other questions come up.

from cadrl_ros.