Awesome
Social-NCE + CrowdNav
Website | Paper | Video | Social NCE + Trajectron | Social NCE + STGCNN
This is an official implementation for
Social NCE: Contrastive Learning of Socially-aware Motion Representations
<br>
<a href="https://sites.google.com/view/yuejiangliu/">Yuejiang Liu</a>,
<a href="https://qiyan98.github.io/">Qi Yan</a>,
<a href="https://people.epfl.ch/alexandre.alahi/?lang=en/">Alexandre Alahi</a>, ICCV 2021
<br>
TL;DR: Contrastive Representation Learning + Negative Data Augmentations 🡲 Robust Neural Motion Models
- Rank in 1st place on the Trajnet++ challenge since November 2020 to present
- Significantly reduce the collision rate of SOTA human trajectroy forecasting models
- SOTA on imitation / reinforcement learning for autonomous navigation in crowds
[New] our more recent work on this topic:
Towards Robust and Adaptive Motion Forecasting: A Causal Representation Perspective, CVPR 2022.
Preparation
Setup environments follwoing the SETUP.md
Training & Evaluation
- Behavioral Cloning (Vanilla)
python imitate.py --contrast_weight=0.0 --gpu python test.py --policy='sail' --circle --model_file=data/output/imitate-baseline-data-0.50/policy_net.pth - Social-NCE + Conventional Negative Sampling (Local)
python imitate.py --contrast_weight=2.0 --contrast_sampling='local' --gpu python test.py --policy='sail' --circle --model_file=data/output/imitate-local-data-0.50-weight-2.0-horizon-4-temperature-0.20-nboundary-0-range-2.00/policy_net.pth - Social-NCE + Safety-driven Negative Sampling (Ours)
python imitate.py --contrast_weight=2.0 --contrast_sampling='event' --gpu python test.py --policy='sail' --circle --model_file=data/output/imitate-event-data-0.50-weight-2.0-horizon-4-temperature-0.20-nboundary-0/policy_net.pth - Method Comparison
bash script/run_vanilla.sh && bash script/run_local.sh && bash script/run_snce.sh python utils/compare.py
Basic Results
Results of behavioral cloning with different methods.
<img src="docs/collision.png" height="240"/> <img src="docs/reward.png" height="240"/>
Averaged results from the 150th to 200th epochs.
<table><tbody> <!-- START TABLE --> <!-- TABLE HEADER --> <th valign="bottom"></th> <th valign="bottom">collision</th> <th valign="bottom">reward</th> <!-- TABLE BODY --> <tr><td align="left">Vanilla</td> <td align="center">12.7% ± 3.8%</td> <td align="center">0.274 ± 0.019</td> <tr><td align="left">Local</td> <td align="center">19.3% ± 4.2%</td> <td align="center">0.240 ± 0.021</td> <tr><td align="left">Ours</td> <td align="center">2.0% ± 0.6%</td> <td align="center">0.331 ± 0.003</td> </tr> </tbody></table>Citation
If you find this code useful for your research, please cite our papers:
@inproceedings{liu2021social,
title={Social nce: Contrastive learning of socially-aware motion representations},
author={Liu, Yuejiang and Yan, Qi and Alahi, Alexandre},
booktitle={Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV)},
pages={15118--15129},
year={2021}
}
@inproceedings{chen2019crowd,
title={Crowd-robot interaction: Crowd-aware robot navigation with attention-based deep reinforcement learning},
author={Chen, Changan and Liu, Yuejiang and Kreiss, Sven and Alahi, Alexandre},
booktitle={International Conference on Robotics and Automation (ICRA)},
pages={6015--6022},
year={2019}
}