Kicking off the new year with a paper accepted to #ICRA2022! In it, we propagate perceptual state uncertainty through trajectory forecasting, making use of a new statistical distance-based loss formulation to do so. Check it out on arXiv: arxiv.org/abs/2110.03267. See you in May!

Can we leverage tools from statistical inference to build safety critical warning systems with a guaranteed ε false negative rate using as few as 1/ε data points? Check out our new paper on sample-efficient safety assurances using conformal prediction! arxiv.org/abs/2109.14082…

Former ASLer, Prof. Navid Azizan has an opening for a postdoctoral scholar in his lab!

Can we verify the safety of a deep neural network for deployment in safety-critical settings? This is a non-convex problem in general, and there have been many existing relaxations constructed for it. 1/2

We unify several SDP relaxations for ReLU neural network verification by providing an exact convex formulation. This provides a path for relaxations that systematically trade off tightness & efficiency proceedings.mlr.press/v151/brown22b/… w/ Robin Brown, Ed Schmerling & @MarcoPavoneSU 2/2

Great collaboration between the Institute for Dynamic Systems and Control and the Automatic Control Lab at ETH Zürich (Nicolas Lanzetti, Andrea Censi, Emilio Frazzoli) and the Stanford ASL at Stanford University (Marco Pavone). Check out the early access version at: lnkd.in/ejPuZrm3

We won the best paper award at the AI4Space Workshop! Here's our framework for how ML models can *detect* and *adapt* to changing input distributions, using OOD detection, subsampling, and continual learning. arxiv.org/abs/2209.06855 #eccv Marco Pavone Stanford ASL The Aerospace Corporation

New on arXiv: we present a learning control approach capable of safe and efficient online adaptation. Our approach combines elements of classical adaptive control, modern robust MPC, and Bayesian meta-learning to yield guaranteed-safe online adaptation! arxiv.org/abs/2212.01371 🧵

Can we learn efficient algorithms to solve classical optimization problems over graphs? In our recent Learning on Graphs Conference 2025 paper, we propose graph reinforcement learning as a general framework to solve network control problems! 📜 openreview.net/forum?id=1sPcf… 🧵👇 (1/n)

Out-of-distribution inputs derail predictions of ML models. How can we cope with OOD data in robotics? How do we even define what makes data OOD? We provide a perspective paper arguing a system-level view of OOD data in robotics! 🧵 (1/5) Now on Arxiv: arxiv.org/abs/2212.14020

Exciting first day co-teaching Marco Pavone’s AA203: Optimal and Learning-Based Control, with Spencer M. Richards at Stanford Engineering! Interested in the intersections between optimal control and RL? Look no further, all course materials will be available at: stanfordasl.github.io/aa203/

Excited to share that our paper on Graph-Reinforcement Learning was accepted at #ICML2023! We present a broadly applicable approach to solve graph-structured MDPs through the combination of RL and classical optimization. Website: sites.google.com/stanford.edu/g… 🧵👇(1/n)+quoted tweet

Excited to present "Learning Control-Oriented Dynamical Structure from Data" next week at #ICML2023! We enforce factorized structure in learned dynamics models to enable performant nonlinear control. Paper: arxiv.org/abs/2302.02529 Code (w/ #JAX): github.com/spenrich/Learn…

📢 Announcing the first Conference on Robot Learning workshop on Out-of-Distribution Generalization in Robotics: Towards Reliable Learning-based Autonomy! #CoRL2023 🎯 How can we build reliable robotic autonomy for the real world? 📅 Short papers due 10/6/23 🌐 tinyurl.com/corl23ood 🧵(1/4)

🔍 How can we detect system-level reasoning failures to improve the robustness of robotic systems in safety-critical settings? We use LLMs as intelligent runtime monitors to reason over and identify potentially problematic elements in a scene! 🧠 tinyurl.com/llm-anomaly

Can we leverage Transformer models to boost trajectory generation for spacecraft rendezvous? In our recent IEEE Aerospace Conference paper, we introduce ART🎨(Autonomous Rendezvous Transformer) to solve complex trajectory optimization problems. Website🌐rendezvoustransformer.github.io A thread 👇

❓ How can we enable real-time reactive control with LLMs for dynamic robotic systems? At #RSS2024 we present AESOP: A 2-stage (🐢🐇) reasoning framework that uses LLMs to increase closed-loop robot trustworthiness in OOD scenarios! Site: tinyurl.com/aesop-llm 🧵(1/6)

Join our ITSC Tutorial on "Data-driven Methods for Network-level Coordination of AMoD Systems Across Scales" tomorrow!

💡For human-robot interaction, human preferences need to be captured at all levels of the robot planning stack: task, motion, and control! Check out Text2Interaction from Jakob and Christopher Agia

🔔Scalable and safe deployment of generative robot policies in the real world requires that we actively monitor their behavior and issue warnings when they are failing. Check out Christopher Agia and Rohan Sinha latest work on runtime monitoring for generative robot policies.