Suning Huang

PhD @Stanford｜BEng @Tsinghua_Uni. Learning to teach robots to learn. Nice to meet you ;)

[Suning Huang @suning_huang]

🤖Low-data post-training can teach a VLA policy a new robot skill. But it also makes it too attached to the training demos.

We call this lock-in🔒: the policy can execute the post-training task, yet fails to respond to seemingly obvious prompt changes.

DeLock preserves steerability using only the policy’s own pretrained knowledge. No extra supervision needed!🚀🚀🚀

#Robotics #AI #EmbodiedAI #VLA

[Ria Doshi @riadoshi21]

🤔 Can we train one VLA policy to control multi-robot teams without any explicit communication?

✨ Introducing CHORUS: a single policy for decentralized, multi-embodiment collaboration

🧵⬇️

[Suning Huang @suning_huang]

🥳Super excited that our paper GRaD-Nav++ has received the RA-L 2025 BEST PAPER!
Huge congratulations to the amazing team🤩 @QianzhongChen, Naixiang Gao, JunEn Low, Timothy Chen, @JiankaiSun, @MacSchwager!

https://t.co/7szi1JeF8i

[Qianzhong Chen @QianzhongChen]

Can't attend #ICRA2026, but happy to share that our work has won RA-L 2025 𝗕𝗘𝗦𝗧 𝗣𝗔𝗣𝗘𝗥

This work explores the alignment between language and action in drone navigation

Thanks my amazing advisor @MacSchwager and coauthors! Thanks IEEE RAS community! Full paper in threads

[Guowei Xu @Kevin_GuoweiXu]

🚀 How should LLMs sample on hard reasoning problems during post-training and inference where direct rollouts rarely produce a correct answer?

Best-of-N (e.g., GRPO) and tree search share two limitations:
🔻 Verification signals are sparse
🔻 Candidates stay within the model's own distribution

We introduce BES: Bidirectional Evolutionary Search — a search framework that couples forward candidate evolution with backward goal decomposition.
✅ Works for both post-training and inference.

[Baiye Cheng @Shutter_Chen]

Great work! Congrats, Suning!

[Robots Digest 🤖 @robotsdigest]

Ever fine-tuned a VLA policy on a small demo dataset and it suddenly stops listening to new instructions?

This paper calls it lock-in. The model just repeats what it saw during training
like always picking bread even when you say apple

Low-data post-training quietly kills steerability

The fix?
DeLock is surprisingly simple and clever

[Suning Huang @suning_huang]

Thanks for the thoughtful point! DeLock is not meant to replace SFT or make arbitrary unseen skills work out of the box. It aims to reduce the combinatorial burden of SFT by leveraging the pretrained backbone to connect post-trained skills with related novel instructions, so we don’t need demos for every variation.

So its effectiveness depends on both the similarity between the trained and novel tasks, and how much the VLA backbone already knows about the relevant concepts/skills.

[Suning Huang @suning_huang]

🤖Low-data post-training can teach a VLA policy a new robot skill. But it also makes it too attached to the training demos.

We call this lock-in🔒: the policy can execute the post-training task, yet fails to respond to seemingly obvious prompt changes.

DeLock preserves steerability using only the policy’s own pretrained knowledge. No extra supervision needed!🚀🚀🚀

#Robotics #AI #EmbodiedAI #VLA

[Suning Huang @suning_huang]

🤖Low-data post-training can teach a VLA policy a new robot skill. But it also makes it too attached to the training demos.

We call this lock-in🔒: the policy can execute the post-training task, yet fails to respond to seemingly obvious prompt changes.

DeLock preserves steerability using only the policy’s own pretrained knowledge. No extra supervision needed!🚀🚀🚀

#Robotics #AI #EmbodiedAI #VLA

[Gu Zhang @Gu__Zhang]

Great work, congrats Suning!

[Ville🤖 @VilleKuosmanen]

We still know so little on how to use the learned representations present in VLMs for VLA training.

Great work @suning_huang and team!

[Mac Schwager @MacSchwager]

How well to VLAs generalize to new prompts after SFT? If you've worked with them, you'll know the answer. The problem is the fine tuning methodology, not the model. Suning has a clever and effective solution that requires no new data, just better SFT and inference methods. 👇

[Yanjiang Guo @Yanjiang_Guo]

I am surprised that so many pre-trained knowledge can be preserved with no additional data if you finetune VLA in a proper way! Check this solid work from Suning!

[Jiankai Sun @JiankaiSun]

Really nice work on tackling “lock-in” in VLA policies!

VLA post-training robustness is a bottleneck, and it’s great to see a method that improves adaptability without extra supervision. DeLock looks like a promising direction.🔥

[Jeannette Bohg @leto__jean]

Ever post-trained a VLA and watched it ignore every novel instruction?

We call this lock-in.

Prior fixes bloat datasets with foundation model labels. 🔓DeLock is different: regularized finetuning + contrastive prompts at inference.

Result: Pretraining priors preserved.

[Qianzhong Chen @QianzhongChen]

I always feel frustrated to see the finetuned VLA policy become useless to any other task. We need generalizable, steerable VLA that can perform well on multiple tasks (all the tasks ultimately). Checkout DeLock that elegantly solve this problem!

[Ruiqian Nai @RuiqianNai]

Recover steering ability of the pre-trained VLA with simple and efficient post training harness👍 Check Suning’s DeLock

[Stanford IPRL Lab @StanfordIPRL]

New work led by @suning_huang exploring how we can preserve steerability during low-data post-training for VLAs 👀

Details below! ⬇️⬇️

[Suning Huang @suning_huang]

💡n/n Huge thanks to my wonderful collaborators @JiaqiShao0819, @ke_wang123, @QianzhongChen, @JiankaiSun, @GYanjiang and amazing advisors @MacSchwager, @leto__jean for all the ideas, feedback and support that made this project possible!

[Suning Huang @suning_huang]

💡6/n More details here:

🌐 Project website: https://t.co/95FBHn83lA
📄 Paper: https://t.co/DLdAqSvlYs
🎥 Video: https://t.co/6J32OPm9Ir