Code for translating navigation instructions in natural language to a high-level plan for behavioral navigation for robot navigation
Switch branches/tags
Nothing to show
Clone or download
Latest commit 981300f Jul 11, 2018
Type Name Latest commit message Commit time
Failed to load latest commit information.
codes corrected starter script Jul 12, 2018
scripts Training model Jul 12, 2018 add ack Jun 19, 2018 corrected starter script Jul 12, 2018
requirements.txt add files Jun 14, 2018

Translating Navigation Instructions in Natural Language to a High-Level Plan for Behavioral Robot Navigation

Xiaoxue Zang*, Ashwini Pokle*, Marynel Vázquez, Kevin Chen, Juan Carlos Niebles, Alvaro Soto, Silvio Savarese

* equal contribution.


Run the following command to install dependencies and libraries, download glove embeddings and dataset for experiments.




python codes/ --experiment_name=[unique name for model folder] --data_dir data

You can also download the Trained model


Evaluate the F1 score and Exact Match on the dev/test dataset, which can be decided by --file_in_path.

python codes/ --train_dir=experiments/[experiment name] --mode=official_eval --data_dir data --file_in_path [dev/test]

--file_out_path is needed if you want to write the prediction out to a file. [--write_out=True]

Run the following instruction to evaluate the result and plot the error analysis.

python codes/ ground_truth_file prediction_file

Show examples

Show the predictions of some examples. --file_in_path decides the dataset to test with. There are three valid arguments: dev, test, test_diff_maps, which respectively mean validation set, test-repeated map set, test-new map set.

python codes/ --train_dir=experiments/[experiment name] --mode=show_examples --data_dir data --file_in_path [dev/test/test_diff_maps] --print_num 10


Example training commmand:

python codes/ --experiment_name=my_experiment --mode=train --data_dir my_data

Example test command:

python codes/ --train_dir experiments/my_experiment --mode=official_eval --data_dir my_data


Toyota Research Institute ("TRI") provided funds to assist the authors with their research but this article solely reflects the opinions and conclusions of its authors and not TRI or any other Toyota entity.

code modified from the Default Final Project for CS224n, Winter 2018