the issuse about sim2real

[2200790303liuyu]

I want to implement this strategy on the real robot, the work I have completed are
1 drawing the SLAM map of the current actual environment
2 extracting the obstacle pixel information for strategy training,
3 getting the training config file,
4 installing the dr_spaam library, and realize the connection between the lidar, robot and the related ros package
Now, when I test it on my robot the problems encountered are:

1. I got the tf information of the map-base_link of the target point through TF, Is this the x,y information that the terminal needs to input
2. After I type the coordinates in the terminal, the mobile robot will not find my end point and will collide with a static obstacle. Can you tell me why this might exist? Thank you so much

[Shuijing725]

It's hard to tell the reason just based on your description. My suggestion is to start from an easier case: To test whether the robot is able to navigate in an empty, open environment, you can set sim2real.use_dummy_detect=True in config.py to remove the detected humans and set the goal to somewhere with no obstacles in between. After passing this test, you can gradually add the obstacles and humans back.

[2200790303liuyu]

Thank you very much for your reply. I will follow your suggestions first, but I think maybe there were somthing wrong with my trainning.
I tried to change the rosTurtlebot2iEnv-v0 to CrowdSim3DTbObs-v0 to test it in the simulation environment, I finded that there has no obstacles in the simulation. The output.gif just has human and robot, but there has no static obstacles.
I have chaned the ax.set_xlim and ax.set_ylim in train.py,and I also changed the sim.arena_size in config.py. It didn't seem to work . I am confused that I can't find the reason now. I think this error might be related to my bot not being able to recognize static obstacles at runtime.
Could you please give me some ideas?Thank you again.

[Shuijing725]

To enable obstacles, in config.py , you need to sim.static_obs = True, and set sim.static_obs_num to a positive value. In the gym environment, the obstacles are stored in a list called self.obstacles.

Before training, I recommend running check_env.py to visualize your simulation environment.

[2200790303liuyu]

Thank you very much for your reply. I will follow your suggestions.

[2200790303liuyu]

Hello author, I am still encountering issues during the process of sim2real.

1. I obtained the coordinates and sizes of obstacles based on my actual scene map.pgm file, and wrote them into the config file,.Now, I have conducted training. Due to my large environment and the coordinate origin not being (0,0), I can see that the entire gray wall model is not in the middle of the screen during running check_dev.py. So, in order to save the complete GIF image, that is, to change the size of the graphics window and coordinate axes, what changes should I make during testing?I have chaned the ax.set_xlim and ax.set_ylim in test.py . It didn't seem to work .
2. I want to know if the coordinates input by the terminal during the sim2real process are coordinate values that are aligned with the trained map coordinate system, or if they are converted coordinates from map --> baselink

[2200790303liuyu]

您好作者，我在 sim2real 的过程中仍然遇到问题。

1. 我根据实际场景 map.pgm 文件获取了障碍物的坐标和大小，并将它们写入配置文件中。现在，我已经进行了培训。由于我的环境很大，坐标原点不是 （0,0），我可以看到在运行_check_dev.py_期间，整个灰色墙模型不在屏幕中间。那么，为了保存完整的 GIF 图像，即更改图形窗口和坐标轴的大小，在测试过程中我应该做哪些更改呢？我已经在 test.py 中唱了 ax.set_xlim 和 ax.set_ylim 。它似乎不起作用。
2. 我想知道终端在 sim2real 过程中输入的坐标是与训练的地图坐标系对齐的坐标值，还是从地图 --> baselink 转换的坐标

The first issue has been resolved. I would like to confirm the second question now. I think it's the coordinates in the global map coordinate system.

[Shuijing725]

The map is only used for robot localization. In navigation, we only need robot odometry and does not need map.

[2200790303liuyu]

Hello author. The model I trained can run normally in the simulation environment. But on my robot, the robot will spin in place and cannot navigate correctly to the target point. When I fix the output action value, for example, I fix its output action1, the robot can accelerate in a straight line until the maximum linear velocity and maintain straight line travel. When I fix the output of action8, the robot can also keep spinning in place. According to this，I think the action the modle output seems to be incorrect. Could you please help me what could be the reason for this.

[Shuijing725]

What robot are you using? Different robots have different dynamics, and this repo is tailored to turtlebot2.

[2200790303liuyu]

Yes, I have noticed this issue. The robot I am using is the Tracer Mini, which is also a two-wheel differential model, so I have always assumed they are the same. I'm not entirely sure what modifications are needed—could you teach me what changes should be made?

[Shuijing725]

You can either replace the turtlebot.urdf with your robot's urdf, or skip the pybullet dynamics and implement your own differential drive model

[2200790303liuyu]

OK, thank you author. I will follow your suggestions.