The parser provided in this package is an implementation of the t5-small parser with language model head for conditional generation.
The semantic parsing task: converting natural language commands -> The Unified Meaning Representation (UMRF). This format powers the TeMoto2.0 packaged developed by the University of Tartu and University of Texas at Austin.
- clone this repository into your
~/catkin_ws/src
- Download the model and tokenizer from slwanna@utexas.edu 's google drive (please contact author of this git repo for access)
- from
~/catkin_ws/src/ROS1_UMRF_T5Parser/scripts
runpip install -r requirements.txt
- Extract the zip in your download then place the
saved_model
andsaved_tokenizer
at the same directory level as theumrf_parser.py
- No
.launch
is included - source your workspace
- start
roscore
in a terminal - navigate to
~/catkin_ws/src/ROS1_UMRF_T5Parser/scripts
python3 umrf_parser.py
This will being running the node. The node listens to natural language commands on the rostopic
: umrf_parses
. The UMRF outputs are published on umrf_parses_output
.
- Provide a RoboFrameNet Parser
- Implement online learning model for live UMRF corrections