Integrate Human prediction with social navigation layer

[Akhil Nagariya]

Hi all,

       I am using the ros navigation stack for some time and have developed a simple prediction algorithm for humans that i wish to integrate with social navigation layer, David Lu suggested me to start the discussion on this group. This is the first time I would be contributing to open source so any help here would be great.

-Akhil     

      

        

[David Lu!!]

Hi Akhil, 

Can you describe what your prediction model does, and how you think it will integrate into a navigation layer? 

-David!!

--
You received this message because you are subscribed to the Google Groups "ros-sig-navigation" group.
To unsubscribe from this group and stop receiving emails from it, send an email to ros-sig-navigat...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Akhil Nagariya]

Hi David,

      I am using hmm to do the long term prediction . User has to provide the possible goals in the environment where a person might go. The algorithm computes the probability that a person follows the direction along the line joining the current position and each of these goals .I was thinking of updating the costmap based on these probabilities.The Implementation has the capability of adding dynamic goals but right now I am not sure how these can be added. I have modified the approach used in Bayesian Human motion Intentionalitiy prediction(BHMIP) : http://www.sciencedirect.com/science/article/pii/S0167865513003176. You can find the code here: https://github.com/akhil22/intention_modelling this is still premature and might need a lot of changes including changes in coding style.  

To unsubscribe from this group and stop receiving emails from it, send an email to ros-sig-navigation+unsub...@googlegroups.com.

[Procópio Stein]

Hi, we are also using pedestrian prediction in our planning. But we are not using move_base. To model the predictions, we create an array of occupancy grids, where each layer represents a time, so we can populate it with the future predictions of pedestrian's positions.
cheers!

[David Lu!!]

Akhil - 

What information does the intention modelling code publish? 

There's a generic "creating a new layer" tutorial here: http://wiki.ros.org/costmap_2d/Tutorials/Creating%20a%20New%20Layer

You can also base new code off of the existing social navigation layer. 

https://github.com/DLu/navigation_layers/tree/indigo/social_navigation_layers

Procópio - I'd like to take a look at your code. Where is it? 

-David!!

To unsubscribe from this group and stop receiving emails from it, send an email to ros-sig-navigat...@googlegroups.com.

[Akhil Nagariya]

Hi David,

Currently it only calculates the probability of different intentions and publishes markers for visualization purposes. I was thinking of updating the costmap based on these probabilities.

I will follow the tutorial and come back to you.

 
On Wednesday, 4 March 2015 22:36:42 UTC+5:30, David!! wrote:

Akhil - 

What information does the intention modelling code publish? 

There's a generic "creating a new layer" tutorial here: http://wiki.ros.org/costmap_2d/Tutorials/Creating%20a%20New%20Layer

You can also base new code off of the existing social navigation layer. 

https://github.com/DLu/navigation_layers/tree/indigo/social_navigation_layers

Procópio - I'd like to take a look at your code. Where is it? 

-David!!

On Tue, Mar 3, 2015 at 2:50 AM, Procópio Stein <procop...@gmail.com> wrote:

Hi, we are also using pedestrian prediction in our planning. But we are not using move_base. To model the predictions, we create an array of occupancy grids, where each layer represents a time, so we can populate it with the future predictions of pedestrian's positions.
cheers!

On Monday, March 2, 2015 at 2:36:49 AM UTC+1, Akhil Nagariya wrote:

Hi David,

      I am using hmm to do the long term prediction . User has to provide the possible goals in the environment where a person might go. The algorithm computes the probability that a person follows the direction along the line joining the current position and each of these goals .I was thinking of updating the costmap based on these probabilities.The Implementation has the capability of adding dynamic goals but right now I am not sure how these can be added. I have modified the approach used in Bayesian Human motion Intentionalitiy prediction(BHMIP) : http://www.sciencedirect.com/science/article/pii/S0167865513003176. You can find the code here: https://github.com/akhil22/intention_modelling this is still premature and might need a lot of changes including changes in coding style.  

On Monday, 2 March 2015 06:25:33 UTC+5:30, David!! wrote:

Hi Akhil, 

Can you describe what your prediction model does, and how you think it will integrate into a navigation layer? 

-David!!

On Sun, Mar 1, 2015 at 5:42 PM, Akhil Nagariya <akhil.n...@gmail.com> wrote:

Hi all,

       I am using the ros navigation stack for some time and have developed a simple prediction algorithm for humans that i wish to integrate with social navigation layer, David Lu suggested me to start the discussion on this group. This is the first time I would be contributing to open source so any help here would be great.

-Akhil     

      

        

--
You received this message because you are subscribed to the Google Groups "ros-sig-navigation" group.

To unsubscribe from this group and stop receiving emails from it, send an email to ros-sig-navigation+unsubscribe@googlegroups.com.

For more options, visit https://groups.google.com/d/optout.