Bayesian Optimization for Learning Nonlinear MPC in Autonomous Agent Navigation
A data-efficient framework that learns nonlinear MPC cost weights via Bayesian Optimization for autonomous agent navigation, deployed on a Unitree Go2 quadruped.
PHYSICAL AI · VISION–LANGUAGE–ACTION
Teaching humanoids to learn skills
and choose the right one.
I'm Lorenzo Ortolani — roboticist, co-founder & CTO of TalOS Robotics AI. I build VLA policies that let a single robot acquire a library of skills and generalize across diverse industrial tasks.
THE MISSION
One robot. Many skills. The right choice.
Industrial robots today are brittle — hard-coded for one job. I work on Vision–Language–Action policies so a single embodiment can learn a growing library of skills and decide which one to deploy for the task in front of it, generalizing across messy, real-world industrial settings.
👁️
Perceive
Multimodal perception — vision, language, and proprioception fused into a shared world model.
🧠
Reason & Choose
The policy reads the task, retrieves the right skill from its library, and sequences sub-skills to solve it.
🦾
Act & Generalize
Closed-loop control that transfers from simulation (MuJoCo, Isaac Sim) to real hardware and unseen tasks.
Professional Summary
I build **Physical AI** — robots that perceive, reason, and act in the real world.
My focus is advancing **Vision-Language-Action (VLA)** policies so that a single
humanoid can learn a diverse library of skills and *choose the right one* for whatever
industrial task it faces, generalizing far beyond a single demonstration.
I co-founded TalOS Robotics AI to put this in operators' hands, and I work across the
full stack: locomotion and navigation, learned manipulation, sim-to-real, and the
agent tooling that wraps it all together.
Education
MSc Automation and Control Engineering — Robotics & ML
2024-09-01
Politecnico di Milano
BSc Automation Engineering
2021-09-01
2024-09-01
Politecnico di Milano
Interests
Physical AI & Embodied Intelligence
Vision-Language-Action (VLA) Policies
Humanoid & Legged Robotics
Reinforcement & Imitation Learning
Autonomous Navigation & Sim-to-Real
FEATURED WORK
Robots in the loop
Navigation, learned manipulation, and sim-to-real — across humanoids, quadrupeds, and drones.
Go2 Autonomous Navigation
Nonlinear MPC tuned by Bayesian Optimization on a Unitree Go2 quadruped. ICRA 2026
<a class="lo-vcard" href="#projects">
<div class="lo-vcard__screen">
<video autoplay muted loop playsinline preload="metadata"><source src="/videos/humanoid-sim.mp4" type="video/mp4" /></video>
</div>
<div class="lo-vcard__body">
<h3>Humanoid VLA in Simulation</h3>
<p>Skill-library policies trained in MuJoCo & Isaac Sim. <span class="lo-tag">Physical AI</span></p>
</div>
</a>
<a class="lo-vcard" href="#projects">
<div class="lo-vcard__screen">
<video autoplay muted loop playsinline preload="metadata"><source src="/videos/moonbot.mp4" type="video/mp4" /></video>
</div>
<div class="lo-vcard__body">
<h3>MoonBot — YC Hackathon</h3>
<p>Edge-first inspection robot with an ACT imitation policy. <span class="lo-tag">w/ Innate</span></p>
</div>
</a>
<a class="lo-vcard" href="#projects">
<div class="lo-vcard__screen">
<video autoplay muted loop playsinline preload="metadata"><source src="/videos/drone-ekf.mp4" type="video/mp4" /></video>
</div>
<div class="lo-vcard__body">
<h3>GPS-Denied Drone Autonomy</h3>
<p>EKF sensor fusion + ORB-SLAM3 + MPC on Jetson Orin / PX4. <span class="lo-tag">AEA</span></p>
</div>
</a>
Research & Publications