Enrico Marchesini

🎯 39+ environments over 7 base grids, including: 🔹 Topology optimization actions 🔹 Redispatch & curtailment operations 🔹 Idle & recovery heuristics 🔹 CMDP-based constraints (overload, islanding, load shedding)

📦 RL2Grid is a benchmark for training & evaluating RL agents on a multitude of realistic grid control tasks, modeling: ✅ AC power flow ✅ Multi-step physics simulation ✅ Operator heuristics ✅ Real-time contingencies ✅ Safety constraints

🚨 Excited to share RL2Grid! A benchmark suite of RL tasks for realistic power grid operations ⚡ Built with TSOs, RL2Grid aims to bring RL closer to real-world critical infrastructure. 📄 Preprint: arxiv.org/pdf/2503.23101 💻 Code:  github.com/emarche/RL2Grid More details below 👇

🚨 Less than 48 hours left to submit to the 17th Adaptive Learning Agent workshop at @AAMASconf! 🚨 We welcome full papers, work in progress, and 2-page abstracts of recent journal papers. Don't miss the deadline! 🔗 More details: ala-workshop.github.io

@c_voelcker It's strange, and we've also started to work on it recently! But according to the first comments, (some) people don't like the idea of being able to reset to arbitrary states.

Question for people working on policy-gradients: if you assume access to a simulator that spits out trillions of samples, shouldn't you also assume (and build in) oracle access to it? It seems strange to me (and maybe only explainable from history) to have one but not the other?

Want to enumerate all the (un)safe regions for a given safety property and Deep Neural Network? Check out the preprint of our new paper accepted at #AAAI24🤩! See you in Vancouver 🇨🇦 arxiv.org/pdf/2308.09842…

Had a great time this week with my co-authors @lmarza_ , @alex_sandrof, @cjdamato ! Here is the poster session for our @aamas2023 work "Safe Deep Reinforcement Learning by Verifying Task-Level Properties"; check it out!

Excited to have presented today our work “Online Safety Property Collection and Refinement for Safe Deep Reinforcement Learning in Mapless Navigation” together with my colleague and friend Enrico Marchesini @_emarche at @ieee_ras_icra 2023! Paper here: arxiv.org/pdf/2302.06695…

Want to replace cost functions in Safe Deep RL? Check out how to employ task-level knowledge to bias policies away from unsafe states at @aamas2023! Congrats to @lmarza_ @alex_sandrof @cjdamato for our work "Safe Deep Reinforcement Learning by Verifying Task-Level Properties"!

Looking for novel ways to improve value approximation and analyzing the effects on Deep Policy Gradient algorithms? Check out our "Improving Deep Policy Gradients with Value Function Search" paper at @iclr_conf with @cjdamato ! iclr.cc/virtual/2023/p…