Chen Tang

@ChenTangMark

Assistant Professor in New Mobility @UCLA | CS Postdoc @UTAustin | ME PhD @UCBerkeley

Los Angeles, CA

chentangmark.github.io

Joined June 2020

653 Following

289 Followers

Replies Reposts Media

Pinned
Chen Tang
@ChenTangMark
Apr 26, 2025
I’m excited to share that I will be joining the CEE Department at @UCLA as an Assistant Professor in New Mobility this Fall! I am recruiting Ph.D. students for Fall25/Spring 26/Fall26. There are also opportunities for master’s, undergrads, and postdocs.
3
Chen Tang
@ChenTangMark
Jul 18
It’s impressive to see how driving-policy training can be scaled with no human demos! Glad to be part of TerraTransfer, where we show how TerraZero’s self-play planner can be distilled into an end-to-end policy with minimal real-world data through representation alignment.
Wei ZHAN
@Wei_ZHAN_
Jul 14
To obtain robust autonomy, it has always been the desire for planners to experience large-scale, high-value scenarios in ultrafast closed-loop training with minimal unit cost of data and compute, ideally self-evolving even without offline human demonstration. TerraZero is our
Chen Tang
@ChenTangMark
Jun 10
RTC is a key ingredient for deploying high-latency VLA policies in real-time. We show that discrete diffusion is a more natural fit for asynchronous execution: with no extra implementation or specialized fine-tuning, it achieve strong performance on dynamic manipulation tasks!
Pengcheng Wang
@Outsider_pc
Jun 9
Real-time Chunking (RTC) is designed to enable smooth asynchronous execution of flow-matching policies. However, it has some critical limitations: its inpainting-based async execution capability comes from inference-time corrections rather than the base policy, yielding little
Chen Tang
@ChenTangMark
Mar 14
Excited to share our new findings on continual RL for VLA models. Surprisingly simple methods can already enable strong performance. Great work led by @JiahengHu1 and @jayjshim
Jiaheng Hu
@JiahengHu1
Mar 13
VLA models are capable generalists. But can they continually self-improve? Such Continual Reinforcement Learning (CRL) problems are traditionally considered very challenging. Surprisingly, we found that with the right setup, the simplest CRL recipe can work really well!
Chen Tang
@ChenTangMark
Sep 28, 2025
How can we align a policy with human preferences efficiently? MEReQ tackles this by inferring residual rewards that capture discrepancies from human experts and adapting via residual Q-learning (arxiv.org/pdf/2306.09526). Check it out at CoRL along with @Yuxin’s other work, ReLIC!
Yuxin Chen
@ThomasYuxinChen
Sep 26, 2025
Two papers to be presented at #CoRL2025! ReLIC: Loco-manipulation with interlimb coordination 🎙️ Oral 1 (Manipulation I)-7, Poster 1-56 🔗 relic-locoman.rai-inst.com MEReQ: Sample-efficient alignment from human intervention 🎙️ Spotlight 2-44, Poster 1-36 🔗 thomaschen98.github.io/mereq.github.i…

Chen Tang

Chen Tang’s posts