| ROS Resources: Documentation | Support | Discussion Forum | Index | Service Status | ros @ Robotics Stack Exchange |
Hello,
As a part of my thesis I want to perform autonomous docking for my robot. I saw that xiaomi vacuum cleaner perform autonomous docking only with LIDAR. I am trying to do the same but can't understand the logic. I bought dock station and have LIDAR. Does anybody do something like this and can help me or give any suggestions?
Thanks in advance, Regards
The xiomi is using reflectors inside the dock
Here is a video I maked to explain it. https://youtu.be/pZJIKH9PdpM
The magic is you getting not only the distances but also something like remission / intensity values from the sensor.
With the Tape you create unique pattern. So you could get the distance only with the remission and the angle and do not need distance.
Yes, I saw your video, and actualy thanks you for that!!! I also broke down a station and saw that.
However, as you know the lidar LaserScan msg provides intensity as well as distance, and I have already tried to check it. The intensity was stable 2.0 with my lidar(
You may close this question if you know how tue dock works
I look at the driver there seems something to bee linke intensity,
@goktuggurler Hi, sorry for the late response. I decided to use lidar which returns not only the distance but the power of returned beam as well. By using that attribute I was testing different reflected materials to achieve the best difference with regular house walls.
But still, the robustness was not perfect. This algorithm probably has to be improved by using not only reflected material but maybe some pattern on the dock station as well. (Like it do Xiaomi vacuum cleaner)
Another way maybe to use more expensive and powerful lidars.
I have a DIY robot that docks itself on the charger. I use AMCL and Navigation stack to put the robot in front of charger and then switch to a camera that can see the charger and drives straight towards it with steering adjustments based location of charger in the camera.
You could also do it using laser feedback to control steering by having a feature at the height of the laser that stands out from the wall behind the charger.Your node then finds the feature closest to the laser and adjust steering to keep that feature straight ahead and move forward.
Additionally you could use mechanical features that force robot and dock into alignment as robot approaches. I haven't needed that but it may help with something like a laser that has lower spacial resolution than a camera.
It really doesn't require perfect odom. In fact it doesn't require encoders at all. It uses the laser to get in front of charger and then alternate (laser or camera) for docking, For Nav Stack laser scan matcher could be used for odom if native encoder based odom is suffering. It does require you know where the charger is though..
Asked: 2019-11-18 07:07:21 -0500
Seen: 1,447 times
Last updated: Dec 28 '20
ROS Answers is licensed under Creative Commons Attribution 3.0 Content on this site is licensed under a Creative Commons Attribution Share Alike 3.0 license.
Maybe some kind of scan matching method? Does your docking station have some unique features that you could detect with the lidar? Maybe use something like RANSAC to fit the lidar data over a model of the docking station.
It is really hard to help you if you don't provide more information about your use-case. If the location of the docking station is static and your environment is not very dynamic you could also use AMCL to just determine your robot pose in a map and define where the docking station is on that map.
Yes, the dock station has some patter, it has like pits. Can you please describe what is RANSAC and how can I scan matching method? Thank you
Only AMCL is not enough((( I want to use AMCL ro reach the region of the station and then perform docking
There are many methods to fit/match laser data to features/models. The way RANSAC (RANdom SAmple Consensus) works is by assuming your docking station is in a certain position, then counting how many of your datapoints (laserpoints) are on top of, or close to, your model, this will give you inliers (points that agree with your guess) and outliers (points that don't agree with your guess). You do this for an "n" amount of guesses and then you pick the one with the greatest amount of inliers (better if you also introduce some threshold where inliers/expected inliers > 0.8 or something). It is a really crude "brute-force" type method, but since you already have an initial guess from AMCL it could work. I'm not saying this is the best solution or that it is easy to implement, but it is 1 way of doing things ...(more)
Additionally, if you are free to change the world/docking station/environment as you please (not sure if this is a constraint in your project) then you could consider adding a feature to the docking station yourself. If you are allowed to add something that is very easily detected by a Lidar, why go through the effort of implementing or coming up with a fancy algorithm? Don't solve a problem that you created! If there is an easier solution go for it!
Thank you for you suggestion, I will try to implement laser/scan matching method and if I fail I will try something else.
Thank you, for the idea)))
Hİ Yehor. How you find any good solution? I have the same project. Using Lidar and reflectors to pose our robot. If possible , write a navigation code to autonomous docking with this pose data.
@bfdmetu Hi, I simply moving with navigation stack to the goal in front of the dock station on the map firstly. And then I am looking for cluster with higher intensity in front of the robot. Because, I have attached reflection material on the dock station which has mostly always higher intensity than other things around the dock station.
1) How can you find higher intensity cluster. How can you use lidar data? I echo topic info /scan but these are meaningless for me :) is there any package or something ?
2) Can you get robot pose info or can you update amcl_pose with this way