RoboPAD · NeurIPS 2026
Beyond Task Success: Post-Training Humanoid Locomotion with Human Preference
RoboPAD: Post-Training Adaptation of Robot Foundation Models (NeurIPS), Paris, France. 2026.
ProceedingsPaper & poster forthcoming
01
Interests
I’m interested in reinforcement learning for real-world problems. Currently I'm exploring robotics and embodied AI, and I'm fine-tuning a SmolVLA at home in my spare time.
Previously, I obtained my PhD at Politecnico di Milano, advised by Marcello Restelli and Francesco Trovò, with a scholarship financed by RSE. I worked on reinforcement learning for cyber-physical systems, specifically water distribution networks under cyber-attacks (during Master's), and on digital twins for battery energy storage systems (2025) within smart grid applications (2026). I'm also the lead author of Gym4ReaL, a benchmark suite with realistic environments for training novel reinforcement learning algorithms (2025).
02
News
- Humanoid locomotion paper accepted at RoboPAD · NeurIPS. Read the paper.
- Defended my PhD on machine learning for battery digital twins.
- Joined the organizing committee for RLSS 2026 in Milan.
- Built the ELLIS Italy website.
- New preprint: Learning in MDPs with Exogenous Dynamics.
Earlier updatesFewer updates
- Presented Gym4ReaL at EWRL in Tübingen.
- Attended MLSS in Arequipa, Peru.
- Two papers at IJCNN. Received a student travel grant and chaired the Reinforcement Learning II session.
- Battery digital twin paper published in Journal of Energy Storage.
- Gave a talk on Deterministic Policy Gradient Methods at Polimi’s AirLab RL3 meetings.
- Started teaching Informatica B at Politecnico di Milano with Prof. Francesco Trovò.
- Placed in the top five at the Air Hockey Challenge. Leaderboard.
- Attended Synapse AI Symposium in Milan.
- Left Moviri to begin my PhD at Politecnico di Milano.
- Presented RL for water systems under cyber-attacks at IFAC CMWRS, Milan.
- Our work on water-system control has been presented at EGU General Assembly, Vienna.
03
Publications
Selected work, from learning theory to real-world applications.
Preprint · 2026
Learning in Markov Decision Processes with Exogenous Dynamics
arXiv:2603.02862. 2026
Preprint · 2025
Gym4ReaL: A Suite for Benchmarking Real-World Reinforcement Learning
arXiv:2507.00257. 2025.
Journal of Energy Storage · 2025
A novel digital twin for battery energy storage systems in micro-grids
Journal of Energy Storage, Elsevier. 2025.
IJCNN · 2025
A Reinforcement Learning Approach for Optimal Control in Microgrids
Proceedings of the IEEE International Joint Conference on Neural Network (IJCNN), Rome, Italy. 2025.
More publications 7
Optimal real-time control of water distribution systems undergoing cyber-physical attacks (Abstract)
EGU General Assembly, Vienna, Austria. 2022.
High-fidelity cyber and physical simulation of water distribution systems. I: Models and Data
Journal of Water Resources Planning and Management, ASCE. 2023.
Gym4ReaL: A Benchmark Suite for Evaluating Reinforcement Learning in Realistic Domains
European Workshop on Reinforcement Learning (EWRL), Tubingen, Germany. 2025.
Optimal Energy Management of Renewable Energy Communities: a Multi-Agent Reinforcement Learning Approach
Machine Learning for Sustainable Power Systems (ML4SPS), ECML-PKDD, Porto, Portugal. 2025.
Leveraging Reinforcement Learning for Micro-grid Optimal Control
Deep Learning Techniques for Observable Smart Grid and Sustainable Energy Systems (DLT-4-OSGE), IJCNN, Rome, Italy. 2025.
Optimal Real Time Control of Water Distribution System undergoing Cyber-Attacks: a Reinforcement Learning approach
2nd IFAC Workshop on Control Methods for Water Resource Systems (CMWRS), Milan, Italy. 2022.
Automatic Control and Reinforcement Learning Techniques for Microgrids
Advances in Digital Twin Computing and Sensor Networks, Elsevier. 2026.
04
Projects & open source
Research tools and personal experiments. Open to feedback and contributions.
Personal project · Robot learning
Training SmolVLA with human demonstrations
From phone recordings to robot demonstrations in LIBERO: motion retargeting, SmolVLA fine-tuning, and analysis of where the learned policy fails.
Open source · Reinforcement learning
Gym4ReaL
A Gymnasium-based benchmark suite for testing reinforcement learning in realistic domains, from energy and water systems to robotic picking.
Open source · Energy systems
ErNESTO Digital Twin
A modular digital twin for battery energy storage, bringing together electrical, thermal, state-of-charge, and aging models for simulation experiments.
05
Talks & seminars
June 2026
Tutoring Session: Basics of Reinforcement Learning I
Reinforcement Learning Summer School (RLSS) · Milan
RLSS 2026 ↗February 2025
Deterministic Policy Gradient Methods
AirLab RL3 PhD meetings · Politecnico di Milano
March 2024
Let’s connect
For research, collaborations, or just to say hello.
davidesalaorni@gmail.comAlso at davide.salaorni@polimi.it