🔥The 9th Round of Easy Loan, Earn $40 Reward is in progress❗️ ⏰ Promotion Period: January 15th - Feburary 15th, 2025 👉 Register now and check more details at gate.io/campaigns/358

Do we need to narrowly redefine "Agent" for LLM-Agents or can we just borrow a broader definition from RL / Embodied AI literature? LLM Agents are agentic in the same sense that a trained robot or an RL policy is agentic. Making this connection more explicit allows us to borrow

I always liked intuitions in the EMNLP'20 contrast sets paper (arxiv.org/pdf/2004.02709). We bring this paradigm to robot policy evaluation! I think there's a deeper space to explore here on contrast sets for sequential decision making in general. Catch us at Conference on Robot Learning :)

When teleoperating a robot, humans often look at the target objects, obstacles or the moving parts of the robot. So their gaze carries saliency information and can help us identify what is important in the task. We show this information improves sample-efficiency in RL and IL.

✨updated paper, blog, code & video of our #EMNLP2024 paper "When Parts Are Greater Than Sums: Individual LLM Components Can Outperform Full Models" w/ Jesse Thomason Robin Jia terarachang.github.io/projects/llm-d…

Thanks to Mark Riedl and other GaTech organizers for inviting me to talk about the work from our GLAMOR lab at the Summit on Responsible Computing, AI, and Society. I am excited about the momentum behind increasing crosstalk between researchers in AI, HCI, policy, and health!

USC research at #CoRL2024! 👏✨ Topics include zero-shot robotic manipulation, comparative language feedback, bimanual manipulation + more 👇 viterbischool.usc.edu/news/2024/11/u… Conference on Robot Learning Yue Wang Erdem Bıyık Daniel Seita 🇺🇦 Jesse Thomason Gaurav Sukhatme USC Viterbi School USC School of Advanced Computing

The CoRL party continues tomorrow! My student Ishika Singh is helping to organize LangRob (sites.google.com/view/langrob-c…) where I'll give the last invited talk at 1600, and she is also presenting our ongoing work on symbolic planning and LLMs (arxiv.org/abs/2406.02791) at LEAP. Come see!

Abrar will be presenting his NVIDIA AI work, ReMEMbR, as a spotlight during LangRob @ CoRL at 0930, and will be at the 1000 poster session on the first floor!

Jesse Thomason gave a really interesting talk where he spent a while talking about how the language we use in robotics is really lame and uninteresting, and most robotics problems are "below" the level of meaningful language

Come say hi! #EMNL2024 this week, featuring research by USC Thomas Lord Department of Computer Science researchers Swabha Swayamdipta Robin Jia Jesse Thomason Sean Ren 🔆 Jaspreet Ranjit and more!✨ USC Viterbi School USC School of Advanced Computing

I missed this post back in JULY when Tanmay made it but it's prescient and even more relevant now. Ccore NLP folks, remember not to re-invent the wheel. Agents are a thing in robotics and reinforcement learning and planning. We have algorithms! Come chat with us!

Excited for Abrar Anwar's work pushing the frontier of active evaluation for robot policies. Big, neural, autoregressive models in other fields get evaluated robustly, but huge eval in robotics is too costly. We need to find the right experiments for the right experimenter cost!

Human-AI decision making is probably something I think is bad but inevitable, so the least we can do is explore ways to reduce inappropriate human reliance on AI system output. In Tejas Srinivasan latest work, he does just that, successfully mitigating /both/ over- and under-reliance.

🌐 Project: liralab.usc.edu/handretrieval/ 📄 Paper: arxiv.org/abs/2505.20455 Amazing collaborators: Matthew Hong , Anthony Liang , Kevin Kim , Harshitha Rajaprakash, Jesse Thomason , Erdem Bıyık Matthew Hong will apply for PhD this year!