Bi3 Official Repository

This repository contains training and evaluation code for the Bi3 dataset. It is a modified version of the AutoBots repository which supports Bi3.

Getting Started

1. Create a python 3.7 environment. I use Miniconda3 and create with conda create --name AutoBots python=3.7
2. Run pip install -r requirements.txt

Setting up and training on Bi3

Convert Bi3 JSON files to trajectory prediction files:

./{path to venv}/bin/python datasets/bi3/create_data_npys.py --bi3-root {path to Bi3 root} --output-root {path to desired output root}/Bi3/trajectory_prediction --overwrite

Training AutoBot-Ego on Bi3:

./{path to venv}/bin/python train.py --exp-id bi3_ego --seed 1 --dataset bi3 --model-type Autobot-Ego --num-modes 6 --hidden-size 128 --num-encoder-layers 2 --num-decoder-layers 2 --dropout 0.1 --entropy-weight 40.0 --kl-weight 20.0 --use-FDEADE-aux-loss True --tx-hidden-size 384 --batch-size 64 --learning-rate 0.00075 --learning-rate-sched 10 20 30 40 50 --dataset-path {path to output root}/Bi3/trajectory_prediction

Training AutoBot-Joint on Bi3:

./{path to venv}/bin/python train.py --exp-id bi3_joint --seed 1 --dataset bi3 --model-type Autobot-Joint --num-modes 6 --hidden-size 128 --num-encoder-layers 2 --num-decoder-layers 2 --dropout 0.1 --entropy-weight 40.0 --kl-weight 20.0 --use-FDEADE-aux-loss True --tx-hidden-size 384 --batch-size 64 --learning-rate 0.00075 --learning-rate-sched 10 20 30 40 50 --dataset-path {path to output root}/Bi3/trajectory_prediction

The same Bi3 Joint defaults can be run from the checked-in config:

./{path to venv}/bin/python train.py --config default_train.yaml

Evaluating an AutoBot model

For the default Bi3 Joint config, evaluate the best-ADE checkpoint with:

./{path to venv}/bin/python evaluate.py --config default_eval.yaml

## Reference

If you use this repository, please cite our work:

@misc{stratton2026bi3dataset, author = {Stratton, Andrew and Singamaneni, Phani Teja and Goyal, Pranav and Alami, Rachid and Mavrogiannis, Christoforos}, title = {{Bi$^3$}: A Biplatform, Bicultural, Biperson Dataset for Social Robot Navigation}, year = {2026}, howpublished = {\url{https://fluentrobotics.com/pdfs/stratton2026bi3dataset.pdf}}, note = {Accessed: 2026-04-30} }