Welcome to SAIL Lab

Safe AI and Robot Learning Lab — School of Computer Science, Shanghai Jiao Tong University

We are a research group working on reinforcement learning, planning, and AI safety, with applications in foundation models (e.g., large language models and multimodal models), robotics, and autonomous systems. Our goal is to design safe, reliable, and efficient learning systems that address pressing real-world challenges — from trustworthy decision-making for embodied agents to scalable and dependable AI. The lab is led by Prof. Shangding Gu (顾尚定).

We are looking for highly self-motivated undergraduate students, graduate (MS/PhD) students, and visiting/intern students who are passionate about reinforcement learning, AI safety, machine learning, large language models, and robotics. If you are interested in our research, please contact us with details about your background and relevant skills.

Join SAIL Lab →

Sponsors

We gratefully acknowledge the support of the following organizations, which enable our research on safe and reliable AI:

Agentic AI Frontier Seminar — an online seminar series on frontier research in agentic AI. Visit the seminar homepage for the speaker lineup, schedule, and past talks.

Seminar Homepage →

Recent News

07.2026: Gave a talk on Agentic AI for Semiconductor Manufacturing at Lam Research.
07.2026: Gave a talk on Agentic AI Scaling at Shanghai AI Lab.
03.2026: Our papers on reinforcement learning for drug design got accepted by Science Advances.
03.2026: Invited to serve as an Area Chair for NeurIPS 2026.
11.2025: Gave a talk on Safe Robot Learning at Shankar Sastry Group, UC Berkeley.
10.2025: Gave a talk on Safe Learning at the Applied Mathematics and Statistics Department Postdoctoral Seminar at Johns Hopkins University.
09.2025: Our papers on safe RL and LLMs got accepted by NeurIPS 2025 and EMNLP 2025.
08.2025: Talk at the 2025 INFORMS Annual Meeting, Atlanta, Georgia, USA (October 26–29).
07.2025: We released Agentic Web, a study on web agent research: Paper | Github.
05.2025: We released M4R, a benchmark for massive multimodal understanding and reasoning in open space (code, dataset and leaderboard available at open-space-reasoning.github.io).
05.2025: We released RLBenchNet, a systematic benchmarking suite for evaluating neural network architectures in reinforcement learning.
04.2025: Gave a talk on Safe Robot Learning at Mac Schwager's Lab, Stanford University.
03.2025: Received a grant from Nvidia to support our research!
03.2025: Lead guest editor for the Special Issue on Trustworthy AI for Automation Control of Embodied Agents in the Era of Foundation Models in IEEE Transactions on Automation Science and Engineering (Call for Papers).
01.2025: Our paper on robust reinforcement learning got accepted by ICLR 2025.
01.2025: We launched the 1st International Workshop on AI Agent Reasoning and Decision-Making (AIR 2025) — workshop homepage.
01.2025: Received a grant from OpenAI to support our research!
01.2025: Our paper on safe multi-objective reinforcement learning got accepted by IEEE Transactions on Pattern Analysis and Machine Intelligence.