Valentino Sacco

Rome, Italy · Jena, Germany

I am a Computer Science PhD Candidate at Sapienza University of Rome, where I am part of the Perception and Intelligence Laboratory (PINlab), supervised by Professor Fabio Galasso. I also work with WarmWind AG in Jena as a Machine Learning Engineer.

My research focuses on Socially Aware Robotics, Human-Robot Interaction, and Embodied AI, with a background in Computer Vision, Deep Reinforcement Learning, and applied machine learning systems.

I am interested in building robots and agents that reason about people, communicate when useful, and act robustly in real-world environments. My current work bridges academic research and production-oriented machine learning engineering across robotics, world models, and embodied intelligence.

news

Jul 01, 2026	Our paper Robots Ask the Way: Communication-Enabled Social Navigation has been accepted at IROS 2026. The paper introduces communication-enabled social navigation, where robots actively ask people for information to locate target individuals in multi-human environments. Read the announcement.
Jun 30, 2026	Our paper Describe-Then-Act: Proactive Agent Steering via Distilled Language-Action World Models has been accepted at ECCV 2026. In this work, we study fast proactive steering for embodied agents through distilled language-action world models. Read the announcement.
May 01, 2026	I am excited to announce that I have started a collaboration with WarmWind AG in Jena as a Machine Learning Engineer, where I will work on applied machine learning and embodied AI systems alongside my PhD research. Read the announcement.
Oct 02, 2025	I am soon entering the third and final year of my PhD studies! After my research visit to Panasonic R&D in Silicon Valley, I am especially excited about opportunities to collaborate with industry alongside my academic work.
Apr 24, 2025	In the next few days I will be attending ICLR 2025 in Singapore, where I will present our spotlight paper: Following the Human Thread in Social Navigation.
Feb 15, 2025	I am visiting Panasonic R&D North America in California for a research stay, where I will also be collaborating with UC Berkeley. I’m thrilled about this opportunity!

selected publications

Describe-Then-Act: Proactive Agent Steering via Distilled Language-Action World Models

Massimiliano Pappa, Luca Romani, Valentino Sacco, Alessio Palma, and 4 more authors

In European Conference on Computer Vision (ECCV), 2026

Abs arXiv Bib

Deploying safety-critical agents requires anticipating the consequences of actions before they are executed. Describe-Then-Act introduces DILLO, a fast steering layer that shifts proactive agent control from visual simulation to semantic outcome prediction via distilled language-action world models. Experiments on MetaWorld and LIBERO show improved policy steering with substantially reduced latency.
@inproceedings{pappa2026describethenact, title = {Describe-Then-Act: Proactive Agent Steering via Distilled Language-Action World Models}, author = {Pappa, Massimiliano and Romani, Luca and Sacco, Valentino and Palma, Alessio and Lathuili\`ere, St\'ephane and Galasso, Fabio and Alameda-Pineda, Xavier and Spinelli, Indro}, booktitle = {European Conference on Computer Vision (ECCV)}, year = {2026}, eprint = {2603.23149}, archiveprefix = {arXiv}, primaryclass = {cs.AI}, url = {https://arxiv.org/abs/2603.23149}, }
Robots Ask the Way: Communication-Enabled Social Navigation

Valentino Sacco, Luca Scofano, Indro Spinelli, and Fabio Galasso

In IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 2026

Abs arXiv Bib

Communication-enabled Social Navigation (CommNav) introduces robotic agents that actively seek assistance from residents to locate target individuals. The work extends Habitat 3.0 with communication-enabled multi-human environments and shows that explicit communication improves multi-person navigation performance while remaining robust to natural, colloquial human language.
@inproceedings{sacco2026robotsasktheway, title = {Robots Ask the Way: Communication-Enabled Social Navigation}, author = {Sacco, Valentino and Scofano, Luca and Spinelli, Indro and Galasso, Fabio}, booktitle = {IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)}, year = {2026}, eprint = {2607.01044}, archiveprefix = {arXiv}, primaryclass = {cs.RO}, url = {https://arxiv.org/abs/2607.01044}, }
Following the Human Thread in Social Navigation

Luca Scofano^*, Alessio Sampieri^*, Tommaso Campari^*, Valentino Sacco^*, and 3 more authors

2025

Abs arXiv Bib Code

The success of collaboration between humans and robots in shared environments relies on the robot’s real-time adaptation to human motion. Specifically, in Social Navigation, the agent should be close enough to assist but ready to back up to let the human move freely, avoiding collisions. Human trajectories emerge as crucial cues in Social Navigation, but they are partially observable from the robot’s egocentric view and computationally complex to process. We present the first Social Dynamics Adaptation model (SDA) based on the robot’s state-action history to infer the social dynamics. We propose a two-stage Reinforcement Learning framework: the first learns to encode the human trajectories into social dynamics and learns a motion policy conditioned on this encoded information, the current status, and the previous action. Here, the trajectories are fully visible, i.e., assumed as privileged information. In the second stage, the trained policy operates without direct access to trajectories. Instead, the model infers the social dynamics solely from the history of previous actions and statuses in real-time. Tested on the novel Habitat 3.0 platform, SDA sets a novel state-of-the-art (SotA) performance in finding and following humans.
@misc{scofano2025followinghumanthreadsocial, title = {Following the Human Thread in Social Navigation}, author = {Scofano, Luca and Sampieri, Alessio and Campari, Tommaso and Sacco, Valentino and Spinelli, Indro and Ballan, Lamberto and Galasso, Fabio}, year = {2025}, eprint = {2404.11327}, archiveprefix = {arXiv}, primaryclass = {cs.RO}, url = {https://arxiv.org/abs/2404.11327}, }