cadrl_ros | ROS package for dynamic obstacle avoidance | Robotics library
kandi X-RAY | cadrl_ros Summary
kandi X-RAY | cadrl_ros Summary
ROS implementation of a dynamic obstacle avoidance algorithm trained with Deep RL. M. Everett, Y. Chen, and J. P. How, "Motion Planning Among Dynamic, Decision-Making Agents with Deep Reinforcement Learning", IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 2018. Paper: Video: The main contribution of this software is the network.py file and trained model parameters (TensorFlow checkpoints). Those contain the policy as reported in our paper and enables other reasearchers to easily compare future algorithms. To make it easy to understand the flow of the code, we provide an example in scripts/ga3c_cadrl_demo.ipynb, in the form of a Jupyter notebook. This can be used just as a reference, but if you want to edit the file, make sure Jupyter is installed in your tensorflow virtualenv to be sure it will work. This assumes you have nvidia-docker installed already. Might work with regular docker with minor changes. That will start an instance of the docker container, and output a Jupyter notebook URL. Copy the URL into a browser, navigate to cadrl_ros/scripts and open ga3c_cadrl_demo.ipynb. Tensorflow and other deps are already installed in the docker container you just built, so the notebook should "just work.". We also provide a ROS implementation that we tested on a Clearpath Jackal ground robot. This node is just one module of the software required for autonomous navigation among dynamic obstacles, and much of it is written as to integrate with our system. The ROS node as written may not be particularly useful for other systems, but should provide an example of how one might connect the modules to test our learned collision avoidance policy on hardware. For example, other systems likely have different representation formats for dynamic obstacles as extracted from their perception system, but it should be straightforward to just replace our cbClusters method with another one, as long as the same information makes it into the state vector when the policy is queried. We recommend looking at the Jupyter notebook first. The algorithm was trained with goals set to be <10m from the agent's start position, so it would be necessary to provide this system with local subgoals if it were to be tested in a long journey. For short distances, say in an open atrium, this is probably not necessary. Clone and build this repo and its dependency (assume destination is ~/catkin_ws/src). Connect inputs/outputs of your system to launch/cadrl_node.launch. As mentioned in the paper, we provide a few datasets that might be useful to researchers hoping to train NNs for collision avoidance. Please find the files in this Dropbox folder, along with instructions for use. The test cases used in the paper are posted in this Dropbox folder. These contain initial positions, goal positions, preferred speed, radius settings for each test case, and are separated by number of agents. They were randomly generated in a way to produce reasonably interesting scenarios to evaluate the algorithms, but since they are random, some may be really easy or boring.
Support
Quality
Security
License
Reuse
Top functions reviewed by kandi - BETA
- This function computes the agent agent
- Update the action
- This is the main function
- Return the observation vector
- Update the state of the agent
- Update the state history
- Check if agent is near the goal
- Convert the state to a vector
- Called when the control is received
- Find the maximum velocity in the vehicle
- Visualize the action
- Handle a frame message
- Visualize a pose
- Called when a message is received
- Find the angle between two angles
- Calculate cluster clusters
- Visualize other agents
- Create the graph
- Create graph output
- Generate a plot of a single action
- Main function
- Plot the current state of the current state in a figure
- Shut down the node
cadrl_ros Key Features
cadrl_ros Examples and Code Snippets
Community Discussions
Trending Discussions on Robotics
QUESTION
I have imported a urdf model from Solidworks using SW2URDF plugin. The model loads correctly on Gazebo but looks weird on RVIZ, even while trying to teleoperate the robot, the revolute joint of the manipulator moves instead of the wheels. Is there anyone who has faced this issue before or has a solution to it? Here is how it looks on Gazebo
Here is the URDF file of the Model:
...ANSWER
Answered 2022-Mar-22 at 13:41So, I realized two problems:
First, you have to change the fixed frame in the global options of RViz to world or provide a transformation between map and world.
Second, your URDF seems broken. There is something wrong with your revolute-typed joints. Changing their type to fixed fixed the problem. I think, it's best if you ask a separate question with a minimal example regarding this second problem.
QUESTION
This question is related to my final project. In gazebo simulation environment, I am trying to detect obstacles' colors and calculate the distance between robot and obstacles. I am currently identifying their colors with the help of OpenCV methods (object with boundary box) but I don't know how can i calculate their distances between robot. I have my robot's position. I will not use stereo. I know the size of the obstacles. Waiting for your suggestions and ideas. Thank you!
My robot's topics :
- cameras/camera/camera_info (Type: sensor_msgs/CameraInfo)
- cameras/camera/image_raw (Type: sensor_msgs/Image)
- sensors/lidars/points (Type: sensor_msgs/PointCloud2)
ANSWER
Answered 2022-Feb-24 at 23:23You can project the point cloud into image space, e.g., with OpenCV (as in here). That way, you can filter all points that are within the bounding box in the image space. Of course, projection errors because of differences between both sensors need to be addressed, e.g., by removing the lower and upper quartile of points regarding the distance to the LiDAR sensor. You can use the remaining points to estimate the distance, eventually.
We have such a system running and it works just fine.
QUESTION
I’m a college student and I’m trying to build an underwater robot with my team.
We plan to use stm32 and RPi. We will put our controller on stm32 and high-level algorithm (like path planning, object detection…) on Rpi. The reason we design it this way is that the controller needs to be calculated fast and high-level algorithms need more overhead.
But later I found out there is tons of package on ROS that support IMU and other attitude sensors. Therefore, I assume many people might build their controller on a board that can run ROS such as RPi.
As far as I know, RPi is slower than stm32 and has less port to connect to sensor and motor which makes me think that Rpi is not a desired place to run a controller.
So I’m wondering if I design it all wrong?
...ANSWER
Answered 2022-Feb-12 at 15:18Robot application could vary so much, the suitable structure shall be very much according to use case, so it is difficult to have a standard answer, I just share my thoughts for your reference.
In general, I think Linux SBC(e.g. RPi) + MCU Controller(e.g. stm32/esp32) is a good solution for many use cases. I personally use RPi + ESP32 for a few robot designs, the reason is,
- Linux is not a good realtime OS, MCU is good at handling time critical tasks, like motor control, IMU filtering;
- Some protection mechnism need to be reliable even when central "brain" hang or whole system running into low voltage;
- MCU is cheaper, smaller and flexible to distribute to any parts inside robot, it also helps our modularized design thinking;
- Many new MCU is actually powerful enough to handle sophisticated tasks and could offload a lot from the central CPU;
QUESTION
In a formation robots are linked with eachother,number of robots in a neighbourhood may vary. If one robot have 5 neighbours how can I find the angle of that one robot with its other neighbour?
...ANSWER
Answered 2021-Dec-15 at 10:03(Following a comment, I replaced the sequence of <face + read heading> with just using towards, wich I had overlooked as an option. For some reason the comment I am referring to has been deleted quickly so I don't know who gave the suggestion, but I read enough of it from the cell notification)
In NetLogo it is often possible to use turtles' heading to know degrees.
Since your agents are linked, a first thought could be to use link-heading, which directly reports the heading in degrees from end1 to end2.
However note that this might not be ideal: using link-heading will work spotlessly only if you are interested in knowing the heading from end1 to end2, which means:
- If your links are directed, it reports the heading from the source to the target;
- If your links are undirected, it reports the heading from the older turtle to the younger turtle.
If that's something that you are interested in, fine. But it might not be so! For example, if you have undirected links and are interested in knowing the angle from turtle 1 to turtle 0, using link-heading will give you the wrong value:
QUESTION
Overlapping targetless stereo camera calibration can be done using feautre matchers in OpenCV and then using the 8-point or 5-point algoriths to estimate the Fundamental/Essential matrix and then use those to further decompose the Rotation and Translation matrices.
How to approach a non-overlapping stereo setup without a target?
Can we use visual odometry (like ORB SLAM) to calculate trajectory of both the cameras (cameras would be rigidly fixed) and then use hand-eye calibration to get the extrinsics? If yes, how can the transformations of each trajectory mapped to the gripper->base transformation and target->camera transformation? Or is there another way to apply this algorithm?
If hand-eye calibration cannot be used, is there any recommendations to achieve targetless non-overlapping stereo camera calibration?
...ANSWER
Answered 2021-Dec-08 at 03:13Hand-eye calibration is enough for your case. Just get the trajectory from each camera by running ORBSLAM. Then, calculate the relative trajectory poses on each trajectory and get extrinsic by SVD. You might need to read some papers to see how to implement this.
This is sometimes called motion-based calibration.
QUESTION
As a premise I must say I am very inexperienced with ROS.
I am trying to publish several ros messages but for every publish that I make I get the "publishing and latching message for 3.0 seconds", which looks like it is blocking for 3 seconds.
I'll leave you with an example of how I am publishing one single message:
...ANSWER
Answered 2021-Nov-29 at 18:44Part of the issue is that rostopic CLI tools are really meant to be helpers for debugging/testing. It has certain limitations that you're seeing now. Unfortunately, you cannot remove that latching for 3 seconds message, even for 1-shot publications. Instead this is a job for an actual ROS node. It can be done in a couple of lines of Python like so:
QUESTION
A c++ novice here! The verbose in the terminal output says the problem is solved successfully, but I am not able to access the solution. What is the problem with the last line?
...ANSWER
Answered 2021-Nov-20 at 02:41You will need to change the line
QUESTION
I'm programming a robot's controller logic. On the controller there is 2 buttons. There is 3 different actions tied to 2 buttons, one occurs when only the first button is being pushed, the second when only the second is pushed, and the third when both are being pushed.
Normally when the user means to hit both buttons they would hit one after another. This has the consequence of executing a incorrect action.
Here is part of the code.
...ANSWER
Answered 2021-Oct-22 at 16:58You could use a short timer, which is restarted every time a button press is triggered. Every time the timer expires, you check all currently pressed buttons. Of course, you will need to select a good timer duration to make it possible to press two buttons "simultaneously" while keeping your application feel responsive.
You can implement a simple timer using a counter in your loop. However, at some point you will be happier with an event based architecture.
QUESTION
I'm trying to put together a programmed robot that can navigate the room by reading instructions off signs (such as bathroom-right). I'm using the AlphaBot2 kit and an RPI 3B+.
the image processing part works well, but for some reason, the MOTION CONTROL doesn't work. I wrote a simple PID controller that "feeds" the motor, but as soon as motors start turning, the robot turns off.
...ANSWER
Answered 2021-Oct-03 at 14:33It is probably not the software. Your power supply is not sufficient or stable enough to power your motors and the Raspberry Pi. It is a very common problem. Either:
- Use separate power supplies which is recommended
- Or Increase your main power supply and use some short of stabilization of power
What power supply and power configuration are you using?
QUESTION
I have read multiple resources that say the InverseKinematics class of Drake toolbox is able to solve IK in two fashions: Single-shot IK and IK trajectory optimization using cubic polynomial trajectories. (Link1 Section 4.1, Link2 Section II.B and II.C)
I have already implemented the single-shot IK for a single instant as shown below and is working, Now how do I go about doing it for a whole trajectory using dircol or something? Any documentation to refer to?
ANSWER
Answered 2021-Oct-16 at 18:09The IK cubic-polynomial is in an outdated version of Drake. You can check out https://github.com/RobotLocomotion/drake/releases/tag/last_sha_with_original_matlab. In the folder drake/matlab/systems/plants@RigidBodyManipulator/inverseKinTraj.m
Community Discussions, Code Snippets contain sources that include Stack Exchange Network
Vulnerabilities
No vulnerabilities reported
Install cadrl_ros
You can use cadrl_ros like any standard Python library. You will need to make sure that you have a development environment consisting of a Python distribution including header files, a compiler, pip, and git installed. Make sure that your pip, setuptools, and wheel are up to date. When using pip it is generally recommended to install packages in a virtual environment to avoid changes to the system.
Support
Reuse Trending Solutions
Find, review, and download reusable Libraries, Code Snippets, Cloud APIs from over 650 million Knowledge Items
Find more librariesStay Updated
Subscribe to our newsletter for trending solutions and developer bootcamps
Share this Page
