Qian Lin (@QianLin44) profile picture

Qian Lin

@QianLin44
Joined May 2026
  • Tweets20
  • Following30
  • Followers5
8Media
Photo by Qian Lin (@QianLin44)Photo by Qian Lin (@QianLin44)Photo by Qian Lin (@QianLin44)Photo by Qian Lin (@QianLin44)Photo by Qian Lin (@QianLin44)Photo by Qian Lin (@QianLin44)Photo by Qian Lin (@QianLin44)Photo by Qian Lin (@QianLin44)
Robot policies are getting remarkably capable—but are they robust enough to deploy? We are moving towards an increasing deployment of pre-trained AI models with publicly released parameters and with no downstream retraining required. But how vulnerable are these pre-trained models to adversarial attacks? Our new work explores White-box, Grey-box and Black-box attacks across a range of both classic and recently proposed Imitation Learning(IL) algorithms, including Vanilla Behavior Cloning (Vanilla BC), LSTM-GMM, Implicit Behavior Cloning (IBC), Diffusion Policy (DP), and Vector-Quantized Behavior Transformer (VQ-BET). 1/3
Image from Akansha Kalra (@KalraAkansha)
2
5
1
9
493
Excited to showcase our paper “ SyncSBC: Decentralized Swarm Behavior Prediction for Synchronized Autonomous Control” at #IROS2026!! 🧵 How does a robot in a swarm know what the collective swarm is doing? Paper: lnkd.in/gmt2CNiJ Website: lnkd.in/gwvUEFVp
Image from Varun Raveendra (@varun_raveendra)
1
4
1
6
331
Super excited to work with an amazing group of people on developing the science of human-robot co-adaptation!
Joydeep Biswas (@Joydeepb_robots) profile pictureJoydeep Biswas@Joydeepb_robotsAug 28
I'm incredibly excited to lead the new NSF Center for Human and Robot Co-Adaptation! It is one of three new $30M, 5yr Centers launched under the @NSF STC program, and is led by @UTAustin , with partners @IUBloomington , @MIT , @TuftsUniversity , @UUtah , and @Yale
4
18
1,383
Excited to share our RLC 2026 paper! Big thanks to @ali_larian for leading, and to Chang Zong Wu & @daniel_s_brown. Teaching agents what you want so it survives environment shift 👇
Ali Larian (@ali_larian) profile pictureAli Larian@ali_larianAug 3
Replying to @ali_larian
Under identical budgets, HSCOT reaches higher constraint coverage and much lower regret on held-out environments than uniform teaching — rewards that survive a change in dynamics. Paper: arxiv.org/abs/2607.08647 Code: github.com/Alilarian/multien… 🧵7/7
Image from Ali Larian (@ali_larian)
1
84
Our paper "A Multi-Layer Sim-to-Real Framework for Gaze-Driven Assistive Neck Exoskeletons" is being presented this week at ICRA 2026! arxiv.org/abs/2603.06779 🧵1/8
Image from Daniel Brown (@daniel_s_brown)
1
4
7
444
What if your task reward is imperfect — especially when it misses safety-related objectives? Maybe crowd preferences can help. Excited to share our paper, Implicit Safety Alignment from Crowd Preferences, accepted to ICML 2026! Grateful to @daniel_s_brown for his guidance and support on this work. Paper: arxiv.org/abs/2605.21822 🧵1/7
Implicit Safety Alignment from Crowd Preferences

Implicit Safety Alignment from Crowd Preferences

Reinforcement Learning from Human Feedback (RLHF) can reveal implicit objectives such as safety considerations that go beyond task completion. In this work, we focus on the common safety criteria...

arxiv.org
1
2
1
4
1,410
more replies
Our approach: compose safe skills instead of combining safety rewards. We first learn preference-aligned skills that capture different user preferences from crowd data. Then, a high-level policy composes these skills to solve new downstream tasks. The key idea is that the high-level policy solves the task, while the low-level skills provide safety 🧵6/7
Image from Qian Lin (@QianLin44)
1
1
32
Across safe RL environments and a proof-of-concept LLM-style setting, our method reduces safety cost without explicit safety rewards, matches oracle safety baselines in task performance, and remains robust to preference imbalance. 🧵7/7
Image from Qian Lin (@QianLin44)
2
5
80
Our approach: compose safe skills instead of combining safety rewards. We first learn preference-aligned skills that capture different user preferences from crowd data. Then, a high-level policy composes these skills to solve new downstream tasks. The key idea is that the high-level policy solves the task, while the low-level skills provide safety 🧵6/7
Image from Qian Lin (@QianLin44)
1
11
Across safe RL environments and a proof-of-concept LLM-style setting, our method reduces safety cost without explicit safety rewards, matches oracle safety baselines in task performance, and remains robust to preference imbalance. 🧵7/7
Image from Qian Lin (@QianLin44)
10
Qian Lin retweeted
Zifan Wu (@zifan_w) profile picture
Zifan Wu@zifan_wMay 18
Excited to share that our paper, "Understanding the Effects of Neuron Dominance in Deep Reinforcement Learning", has been published in Transactions on Machine Learning Research! Full paper: openreview.net/forum?id=VNV1… Work with Qian Lin, Blake Lawlor, Haijun Zhao, and Daniel Brown.
2
1
10
2,357
Super excited about our lab's new paper led by @zifan_w: "Understanding the Effects of Neuron Dominance in Deep Reinforcement Learning", that has been published in Transactions on Machine Learning Research (TMLR)! openreview.net/forum?id=VNV1… 1/N
Zifan Wu (@zifan_w) profile pictureZifan Wu@zifan_wMay 18
Excited to share that our paper, "Understanding the Effects of Neuron Dominance in Deep Reinforcement Learning", has been published in Transactions on Machine Learning Research! Full paper: openreview.net/forum?id=VNV1… Work with Qian Lin, Blake Lawlor, Haijun Zhao, and Daniel Brown.
1
6
13
1,978
Load more