We introduce ACTSEL. A method for automatic selection of actions that help optimally determine physical object properties that are not readily available through vision.
General overview of the algorithm (left), Bayesian network and action relations (right)
For best experience install conda environment as
numpy
,scipy
andscikit-learn
are needed for algorithm operation
- To run the model, fill in the templates for nodes, actions and their relevant confusion matrices in
configs/templates
. In order to update the actual config.json
files, run thescripts/templates_to_cfgs.py
from root directory as:
python3 scripts/templates_to_cfgs.py
- Customize the
main.py
to meet your action and object requirements byt customizingexperiment_object_names
and action to node mapping.
The algorithm and results presented in the paper were obtained offline on pre-measured dataset for broader statistical understanding. This fact is reflected in main.py
.