Jie Chen

Ph.D. student, National University of Singapore

MARMot Lab, advised by Asst. Prof. Guillaume Sartoretti

A*STAR IAIC, AGS scholar, advised by Dr. Wei-Yun Yau

Generalizable robot learning, world models, and foundation models for embodied agents.

About

I am currently a Ph.D. student at the National University of Singapore (NUS). I research in the Multi-Agent Robotic Motion (MARMot) Laboratory advised by Asst. Prof. Guillaume Sartoretti. I’m also an AGS scholar co-supervised by Dr. Wei-Yun Yau at the A*STAR Institute of Advanced Intelligence and Computing (IAIC). Before this, I obtained my Master of Science degree from NUS and a B.Eng degree in Mechatronics from Chongqing University.

My research interests span generalizable robot learning, world models, and foundation models for embodied agents.

Outside the lab

I work out (bench press 100 kg, 25 pull-ups, strict press 70 kg), play the drums (noob level), and listen to a lot of Midwest emo (Chinese/American/Japanese Football, toe, …). When I game, it's usually Overwatch, Minecraft, or The Legend of Zelda.

I also love birds, owls most of all.

Watch the birds

Burrowing owlNational Aviary, Pittsburgh

Toco toucanNational Aviary, Pittsburgh

African penguinsNational Aviary, Pittsburgh

Victoria crowned pigeonBird Paradise, Singapore

TuracoBird Paradise, Singapore

EgretsTaihu Lake, Wuxi

GullsFisherman's Wharf, San Francisco

Black swans and cygnetsChongqing University

Black swans and ducks at Tianjin University — Black swans and ducksTianjin University

9 sightings from 6 places

Publications

MiniWAM: Learning Compact Future Targets for Efficient World-Action Modeling

Jie Chen, Ruofei Bai, Yuxin Cai, Yifeng Zhang, Chengyang He, Jun Li, Wei-Yun Yau, and Guillaume Sartoretti

arXiv 2026arXiv preprint arXiv:2610.12194

Project page arXiv

Instead of predicting the future in a huge native visual feature space, MiniWAM predicts compact PRISM targets learned with inverse dynamics and feature reconstruction. The target uses 65× fewer future tokens, trains up to 8× faster, and beats native-feature prediction; at 0.25B parameters, MiniWAM is competitive with 6B-scale world action models.
```
@article{chen2026miniwam,
  author = {Chen, Jie and Bai, Ruofei and Cai, Yuxin and Zhang, Yifeng and He, Chengyang and Li, Jun and Yau, Wei-Yun and Sartoretti, Guillaume},
  title = {MiniWAM: Learning Compact Future Targets for Efficient World-Action Modeling},
  journal = {arXiv preprint arXiv:2610.12194},
  year = {2026},
}
```
Robots as Tokens: Unified Diffusion Transformer for Coordinated Multi-Robot Trajectory Generation

Ruofei Bai, Jie Chen, Yuxin Cai, Jun Li, Wei-Yun Yau, and Lihua Xie

CoRL 20262026 Conference on Robot Learning (CoRL)

Project page arXiv

Roken treats each robot as a token in a diffusion transformer and generates coordinated trajectories for the whole team in a single feed-forward pass, satisfying both safety and connectivity constraints. One model handles single-robot, multi-robot and conditional planning, scales across team sizes, and outperforms the planner that generated its training data.
```
@inproceedings{bai2026robots,
  author = {Bai, Ruofei and Chen, Jie and Cai, Yuxin and Li, Jun and Yau, Wei-Yun and Xie, Lihua},
  title = {Robots as Tokens: Unified Diffusion Transformer for Coordinated Multi-Robot Trajectory Generation},
  booktitle = {2026 Conference on Robot Learning (CoRL)},
  year = {2026},
}
```
ImagiNav: Scalable Embodied Navigation Via Generative Visual Prediction and Inverse Dynamics

Jie Chen, Yuxin Cai, Yizhuo Wang, Ruofei Bai, Yuhong Cao, Jun Li, Wei-Yun Yau, and Guillaume Sartoretti

IROS 20262026 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)

Project page arXiv

Sidesteps the limited diversity and realism of simulation data through navigation by imagination: a fine-tuned video model imagines the egocentric video of the route, and an inverse dynamics model turns that video into a trajectory. Learning from in-the-wild egocentric human videos, it transfers zero-shot to robots without any robot demonstrations.
```
@inproceedings{chen2026imaginav,
  author = {Chen, Jie and Cai, Yuxin and Wang, Yizhuo and Bai, Ruofei and Cao, Yuhong and Li, Jun and Yau, Wei-Yun and Sartoretti, Guillaume},
  title = {ImagiNav: Scalable Embodied Navigation Via Generative Visual Prediction and Inverse Dynamics},
  booktitle = {2026 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)},
  year = {2026},
  organization = {IEEE},
}
```
Beyond Imitation: Reinforcement Learning Fine-Tuning for Adaptive Diffusion Navigation Policies

Junhe Sheng, Ruofei Bai, Kuan Xu, Ruimeng Liu, Jie Chen, Shenghai Yuan, Wei-Yun Yau, and Lihua Xie

IROS 20262026 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)

arXiv

Diffusion navigation policies learned by imitation are capped by their dataset’s coverage, and in unseen settings distribution shift leads to compounding errors and collisions. Fine-tuning them with GRPO, which needs no value network, while updating only the upper decoder layers and action head over a frozen visual encoder, raises success on unseen Isaac Sim scenes from 52.0% to 58.7% with fewer collisions; the fine-tuned policy also transfers zero-shot to a real quadruped.
```
@inproceedings{sheng2026beyond,
  author = {Sheng, Junhe and Bai, Ruofei and Xu, Kuan and Liu, Ruimeng and Chen, Jie and Yuan, Shenghai and Yau, Wei-Yun and Xie, Lihua},
  title = {Beyond Imitation: Reinforcement Learning Fine-Tuning for Adaptive Diffusion Navigation Policies},
  booktitle = {2026 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)},
  year = {2026},
  organization = {IEEE},
}
```
FISS+: Efficient and Focused Trajectory Generation and Refinement Using Fast Iterative Search and Sampling Strategy

Shuo Sun, Jie Chen, Jiawei Sun, Chengran Yuan, Yuanchen Li, Tangyike Zhang, and Marcelo H. Ang Jr.

IROS 20232023 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)

Paper Code

A coarse-to-fine sampling planner for autonomous driving: a discrete coarse search finds promising regions, then a continuous fine search refines the trajectory there, concentrating new samples in low-cost areas. On the CommonRoad benchmark it beats baseline planners in both trajectory optimality and computation time.
```
@inproceedings{sun2023fiss+,
  author = {Sun, Shuo and Chen, Jie and Sun, Jiawei and Yuan, Chengran and Li, Yuanchen and Zhang, Tangyike and Ang Jr., Marcelo H.},
  title = {FISS+: Efficient and Focused Trajectory Generation and Refinement Using Fast Iterative Search and Sampling Strategy},
  booktitle = {2023 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)},
  year = {2023},
  organization = {IEEE},
}
```

Jie Chen

About

Publications

MiniWAM: Learning Compact Future Targets for Efficient World-Action Modeling

Robots as Tokens: Unified Diffusion Transformer for Coordinated Multi-Robot Trajectory Generation

ImagiNav: Scalable Embodied Navigation Via Generative Visual Prediction and Inverse Dynamics

Beyond Imitation: Reinforcement Learning Fine-Tuning for Adaptive Diffusion Navigation Policies

FISS+: Efficient and Focused Trajectory Generation and Refinement Using Fast Iterative Search and Sampling Strategy