robodhruv / drive-any-robot Goto Github PK

I use default model which you offer ,and CARLA's image is 800600
and in the models.yaml
image_size: [85, 64] # [widht, height] of input images
should i set image_size to [800,600] ?
the error with 800600

the error with 85*64

thanks for your reply

Dear authors,

Thank you for the amazing work and releasing the datasets! I have a question regarding the RECON dataset. In the process_bags.py script the sample rate is set by default to 4 Hz. In the process_recon.py script I didn't find anything related to the sample rate. Do you assume that the sample rate in the RECON dataset is also equal to 4 Hz?

Looking forward for your reply!

Hi,

Nice work! I've been testing the method, and ran into a bug in the deployment version of the image_transform function (https://github.com/PrieureDeSion/drive-any-robot/blob/main/deployment/src/utils.py#L89).

The order of image_size is flipped twice (lines 99, 103), resulting in the image_size being given to resize as WxH instead of HxW. In my experience this was affecting collision avoidance performance to some extent.

Br,
Lauri

Hello, your work is excellent, thank you for sharing. When I tested the GNM model in gazebo, I found that the model did not implement obstacle avoidance. Does the model not have this capability during navigation? Or is it limited obstacle avoidance?

Hello author, I'm very interested in your research, thanks for your great job.

I see that GNM can be deployed in a simulation environment and I would like to ask you how to deploy it to the simulated environments
?

Hello author, I'm very interested in your research, thanks for your great job
and I want to know when the code and data will go to open source？

Hi @PrieureDeSion,
As far as I know, rosbag should record image and odometry topics for deployment step. However, the following line only records the image topic. Is there anything I misunderstood?
https://github.com/PrieureDeSion/drive-any-robot/blob/905c7e3e304157f64e33e6a5aaa92e76aefc98d7/deployment/src/record_bag.sh#L25

Hi

Thank you for sharing great work! In the paper, I found that you used Dijkstra’s algorithm to compute the optimal sequence. However, I could not find any code related to Dijkstra’s algorithm in 'deployment' directory.

Please let me know if I missed anything.

Hello author,

Why do larger datasets have fewer training points,such as scand?

Hi @PrieureDeSion,
Is it enough to prepare image and odometry topic in rosbag file? And is the accuracy of the odometry an important factor in graph generation?

Hi, i try to deploy it with carla,and i collect the .bag file, and i use
./create_topomap.sh topomap_test_name /home/gahho/code/drive-any-robot/deployment/topomaps/bags/test202301191036_2023-01-20-10-53-19.bag
it occurs some error
and i use default
python create_topomap.py
it occurs the same error

thanks for your reply

What is the reason of doing two forward passed through the model during training to calculate losses for distance and action? Is it just implementation detail or there is some reason to do it this way related to propagating loss backward?

https://github.com/PrieureDeSion/drive-any-robot/blob/b58b2d75d4153f28f3c0ce9a68d291822b9cf263/train/gnm_train/training/train_utils.py#L239
https://github.com/PrieureDeSion/drive-any-robot/blob/b58b2d75d4153f28f3c0ce9a68d291822b9cf263/train/gnm_train/training/train_utils.py#L246

I use carla to deploy gnm，the image which rosbag's camera：

the image ./create_topomap.sh create

It looks different,that's right?
Thanks for you reply

There are many types of cameras in the robotics domain. In case of different camera is used in the training and deployment, is GNM work well? Is there any reason to recommend a wide-angle RGB camera in deployment?

Hi, I really appreciate your excellent work !!!!!
I have the following problems while reproducing the project:

1. Dataset: Is the "Berkeley [34] Jackal 2m/s 4h suburban" dataset mentioned in TableI of the paper available for download? Where can I get this dataset？

2. Is it wrong that the platform is ATV in "NeBula [40] ATV 10m/s 10h off-road" mentioned in TableI of the paper? I checked the original NeBula paper (https://arxiv.org/pdf/2103.11470.pdf), and don't see any ATV robots in the paper. I would like to know which dataset is used in the GNM project?

Thanks a lot!!

Hi
thanks for the great work ！
I tried to run your work in the CARLA simulator, but encountered some problems.
First of all, whether I use lagre model or medium model, the car always turns in circles. What is the reason? Is the initial point of the car must be consistent with the initial point recorded by rosbag?
In addition, I found that the vehicle will not avoid obstacles when it encounters obstacles. Is it because this work does not add obstacle avoidance algorithms? Just pure path planning?
Finally, can you provide more information on how your work was run in the CARLA simulator?
Looking forward to your reply, thank you!

Is there documentation for a successful deployment in carla?

Hi, I very much enjoyed reading your paper.

In the paper you write that both the code and data will be made publicy available. I wanted to ask when you plan to do this or, if it is not cleaned yet, if it would be possible to already make them available privately to other researchers like myself in the meantime? That would be fantastic.

Thanks for your great job, i want to deploy it with CARLA,can you offer some document about it?