pierrelux's profile - activity

I had changed the group to "dialout" but was still getting the error. Your answer then fixed the problem. The port is indeed opened twice ! I wondered if someone submitted a patch.

I'm not exactly sure what you're trying to accomplish but I'd like to propose to take a look at the built-in "dir()" function in python. For example:

>>> from sensor_msgs.msg import LaserScan
>>> dir(LaserScan)
['__class__', '__delattr__', '__doc__', '__eq__', '__format__', '__getattribute__', '__getstate__', '__hash__', '__init__', '__module__', '__new__', '__reduce__', '__reduce_ex__', '__repr__', '__setattr__', '__setstate__', '__sizeof__', '__slots__', '__str__', '__subclasshook__', '_check_types', '_connection_header', '_full_text', '_get_types', '_has_header', '_md5sum', '_slot_types', '_type', 'angle_increment', 'angle_max', 'angle_min', 'deserialize', 'deserialize_numpy', 'header', 'intensities', 'range_max', 'range_min', 'ranges', 'scan_time', 'serialize', 'serialize_numpy', 'time_increment']

This would give you the ability to "access all message fields recursively".

In large projects, your code might depend on many external packages. A manifest contain meta-information about what your software is doing and how to set it up. The documentation is quite exhaustive on the topic: wiki.ros.org/Manifest

Practically speaking, you can very well just "import roslib; import rospy" and your code will still work. You might however want to think of writing a manifest as soon as you start depending on other libraries and want to share you code with other people.

Are you sure that the "serial" package that you import is indeed the one you need ? There might be some conflict with other packages. You can try "import serial; help(serial)" to get more details.

Thanks for the suggestions. In terms of sensors, I'd be pretty basic: laser, odometry, no camera.

Thanks for the suggestions. In terms of sensors, I'd be pretty basic: laser, odometry, no camera.

At the abstract level, I'd want to simulator to take an initial state and policy and return the trajectory that the system followed. I would almost use it as a "local planner".

Ideally, I could simulate multiple episodes in parallel across different processes. I can see this as a being a problem due to the client/server architecture.

It seems that Gazebo (wiki.ros.org/simulator_gazebo/Tutorials/RunningSimulatorHeadless) Stage (ros.informatik.uni-freiburg.de/roswiki/stage.html) and STDR (wiki.ros.org/stdr_simulator) can run in headless mode.

I'm looking to sample, as fast as possible, the outcome of some policy for a mobile robot.

This is a reinforcement learning project and it would be prohibitive to have a episode run for say 30 secs.

The presence of noise is preferable but physical realism is not required. Ideally, I would want to deploy my experiment over Amazon EC2. At the abstract level, I'd want to simulator to take an initial state and policy and return the trajectory that the system followed. I would almost use it as a "local planner".

Ideally, I could simulate multiple episodes in parallel across different processes. I can see this as a being a problem due to the client/server architecture.

What would you recommend among Gazebo, Stage or STDR ?

Would OpenRAVE be a better fit ?