tmhoward / h2sl Goto Github PK

Transfer the NL interface. Not super important right now.

Forgot to add newly transferred features to xml file, I will add them

want choice of whether project builds static and/or dynamic libraries

• Transfer functions from adcg into h2sl that are not present.
• Lcm functions are not to be transferred.
• the changes in the demo program also to be transferred.

Want to have a protected member that is a pointer to the parent phrase class.

Integrate ADCG model into h2sl project.

-Create Demo program for each class
-Add dummy data
-Add class_name() function
-fix any issues with class functions

Add demo program to test the class functions for object.cc

Anticipation is that the to/from xml functions are not capturing the position/orientation information.

New branch created for ongoing work on adding spatial regions, exploration to guide grounding of unknown symbols. This supports work by Nicole Glabinski (in Nick Roy's group).

Transfer code to re-create the results for the RSS/IJRR paper. Account for the new models and navigation domain.

Target files:

• test_generator.cc
• test_combine_results.cc
• test_process.cc
• test_convert_xml_to_mat.cc

For example, "the rightmost block" will now have an index symbol expressed as well as an object_property symbol.

Update the demo program to match the abstract_container demo; add a script to run the demo.

Generate a grammar file for the new door_dataset

Branch to cleanly add an initial push of the navigation dataset.

Add the spatial features left.

Copy constructor has small error

Make sure the feature_set.xml file has all instances of all features we need. Also, ensure we are not duplicating features by including specialized an templated features that have the same value function.

• Incorporate new changes from factor_set.cc into factor_set_adcg.cc
• Implement abstract concepts with symbol dict.

The h2sl feature_set.xml has the following:

    <feature_matches_child_object invert="0"/>
    <feature_matches_child_object invert="1"/>
    <feature_matches_child_region invert="0"/>
    <feature_matches_child_region invert="1"/>
    <feature_matches_child_constraint invert="0"/>
    <feature_matches_child_constraint invert="1"/>

These are now templated:
feature_object_matches_child.h and .ipp
The individual feature_matches_child_region etc. have now gone away but still live in the feature_set.xml file.

How do we load in templated features from the feature_set.xml file. Do we have an example?
Jake could you discuss this with Tom and let us know.

Thanks

-add num_phrases() and num_words() functions.
-add to_tikz function

Abstract out the search space class to encapsulate vector of groundings and CVs.

feature_abstract_container_type
feature_abstract_container_color
feature_abstract_container_number
feature_abstract_container_index

Extending the symbol space of h2sl based on h2sl_nsf_nri_mvli

move the examples in the corpus from h2sl_nsf_nri_mvli
create another folder containing examples with no actions

feature_object_matches_child
feature_object_matches_child_container_object
feature_object_merge_abstract_container_index_region_container
feature_container_merge_abstract_container_region_container
feature_container_merge_container_pair
feature_object_merge_container_pair
feature_object_merge_abstract_container_region_container
feature_object_merge_object_property_spatial_relation
feature_object_merge_abstract_container_spatial_relation
feature_object_merge_object_container
feature_object_merge_single_object_different_container
feature_object_merge_object_different_container
feature_object_merge_object_property_container
feature_object_merge_object_property_region_container
feature_object_merge_object_region_container
feature_object_type_matches_child_abstract_container

feature_region_container_type
feature_region_container_container_type

-update abstract_container to have a "class_name" function
-generate a demo program to test the class functions
-fix any issues uncovered during testing

Hi There!

Can the 'adcg' branch reproduce results from your RSS paper (http://www.roboticsproceedings.org/rss12/p37.html)?

Regards

Add template-based features to simplify derived models

feature_object_property_type
feature_object_property_relation_type

Create a demo program and dummy data to test the container.cc functions; add a script to run the program.

Add a demo program to test the class functions for number.cc.

Cleanup and restructure the data folder. Will also add more navigation example data.

Introduce hand-crafted example files for testing in the navigation domain (doors, buildings).

ToDo:

• DCG
• Factor Set
• Inference

• Modify the data structures storing spatial properties in the world model. Go from vectors to hash.
• These are used in the spatial features

Update the data structures for pre-computed data structures for the spatial characteristics. Change from arrays to maps.

This will lead to changes in the spatial features

Add value functions to each feature that allows for compatibility with the monologue context function argument.

Transfer the LLM training process and the arguments file.

To be done after doing DCG.

Add the symbol dictionary in inference functions.