About
I am a PhD candidate at EPFL, Switzerland, working with Prof. Alexandre Alahi at the VITA Lab. Before EPFL I received my M.Sc. in Robotics, Systems and Control from ETH Zurich and B.E. in Navigation Engineering from Wuhan University.
During my master's I worked on human-to-robot handover at the AIT Lab at ETH Zurich with Prof. Otmar Hilliges, in collaboration with NVIDIA. Earlier I worked on generative traffic simulation with Prof. Bolei Zhou at UCLA.
My current research sits at the intersection of data-centric ML, LLM agents & post-training, world models & video generation, and autonomous driving & robotics. Recent work includes PTD (few-step video generation, from my internship at NVIDIA Research), RAP (scalable 3D rasterization for end-to-end driving, powering the 1st-place Waymo 2025 entry), TAROT (optimal-transport data selection, ICML 2025), and Weak-for-Strong (training a small meta-agent to orchestrate frontier LLMs).
I am actively seeking research roles and collaborations on data-centric post-training, agent evaluation, and world models for robotics/AD. Please reach out if our interests overlap.
Selected Publications
First-author or core contributor unless otherwise noted. Full list on Google Scholar.
Lan Feng, Peter Karkus, Maximilian Igl, Julius Berner, Yuxiao Chen, Shuhan Tan, Alexandre Alahi, Boris Ivanovic, Marco Pavone
arXiv 2026 · w/ NVIDIA Research
Few-step distillation for large video diffusion and flow models: the student fits each denoising segment with a polynomial curve under a single least-squares loss and keeps the teacher's architecture. Four steps set a new trajectory-distillation state of the art on Wan2.1-14B, and LoRA-only distillation of the 33B audio-video MiniMax-H3 keeps more motion and diversity than DMD-style methods.
Lan Feng, Wuyang Li, Éloi Zablocki, Matthieu Cord, Alexandre Alahi
arXiv 2026
A one-step image generator trained by matching feature distributions under frozen encoders — no online teacher, adversary, trajectory, or JVP. Sets the one-step ImageNet state of the art, and post-trains four-step FLUX.2 [klein] into a single step that surpasses its teacher.
Lan Feng, Yang Gao, Éloi Zablocki, Quanyi Li, Wuyang Li, Sichao Liu, Matthieu Cord, Alexandre Alahi
ICLR 2026
Scalable data augmentation for end-to-end driving via lightweight 3D rasterization — recovery-oriented perturbation and cross-agent view synthesis at >500k samples. Leads NAVSIM v1/v2 and Bench2Drive leaderboards.
Lan Feng, Fan Nie, Yuejiang Liu, Alexandre Alahi
ICML 2025
Optimal-transport framework for selecting training data aligned with a target distribution. More data is not all you need — the right subset beats larger random ones.
Fan Nie, Lan Feng, Haotian Ye, Weixin Liang, Pan Lu, Huaxiu Yao, Alexandre Alahi, James Zou
COLM 2025 · w/ Stanford
A small trained meta-agent orchestrates stronger frontier executor models for complex reasoning — specialized planners can unlock capabilities without fine-tuning the frontier model itself.
Lan Feng, Mohammadhossein Bahari, Kaouther Messaoud Ben Amor, Éloi Zablocki, Matthieu Cord, Alexandre Alahi
ECCV 2024
Unifies data formats and models across Waymo, Argoverse, and nuScenes to scale trajectory prediction research; widely adopted by the community.
Earlier Work
Lan Feng, Sammy Christen, Wei Yang, Yu-Wei Chao, Otmar Hilliges, Jie Song
ICRA 2024 · w/ ETH AIT Lab & NVIDIA
Yang Gao, Po-Chien Luan, Kaouther Messaoud, Lan Feng, Alexandre Alahi
arXiv preprint, 2025
Lan Feng, Quanyi Li, Zhenghao Peng, Shuhan Tan, Bolei Zhou
ICRA 2023 · w/ UCLA
Quanyi Li, Zhenghao Peng, Lan Feng, Zhizheng Liu, Chenda Duan, Wenjie Mo, Bolei Zhou
NeurIPS 2023
Quanyi Li, Zhenghao Peng, Lan Feng, Qihang Zhang, Zhenghai Xue, Bolei Zhou
IEEE TPAMI, 2022
Quanyi Li, Zhenghao Peng, Haibin Wu, Lan Feng, Bolei Zhou
NeurIPS 2022
Contact
The best way to reach me is by email at lan.feng@epfl.ch. I am based in Lausanne, Switzerland. I also use X/Twitter and LinkedIn.
I am especially interested in collaborating on: (i) data curation for LLM post-training, (ii) long-horizon agent evaluation, and (iii) controllable world models for robotics and autonomous driving. If any of these resonate, drop me a line.
