Pietro Noah Crestaz

Pietro Noah Crestaz

PhD Researcher in Learning for Robotics

LAAS-CNRS · University of Trento · Toulouse, France

Reinforcement Learning Sampling-Based Control World Models Robot Foundation Models

About

PhD candidate at LAAS-CNRS (Toulouse) and the University of Trento, advised by Nicolas Mansard and Andrea Del Prete. My research is on robot learning, in particular world models from state and vision for model-based RL and planning, and universal policies that generalize across tasks and robots. I hold an MSc in Mechatronic Engineering from Politecnico di Torino (with an Erasmus+ year at Aalto University) and a BSc in Industrial Engineering from Trento.

Publications & Preprints

Journal Articles

TD-CD-MPPI: Temporal-Difference Constraint-Discounted Model Predictive Path Integral Control

Project page PDF
Preprints & Under Review

VAMPS: Visual and Motor Policies from Sampling-Based Planning

arXiv

Temporal Difference Learning of Probabilistic Reach-Avoid Value Functions

Project page arXiv PDF

CAST: Alternating State-Value Targets and Expanded Policy Gradients for Model-Based Reinforcement Learning

Project page arXiv PDF

CACTO-BIC: Scalable Actor-Critic Learning via Biased Sampling and GPU-Accelerated Trajectory Optimization

arXiv

COSMIK-MPPI: Scaling Constrained Model Predictive Control to Collision Avoidance in Close-Proximity Dynamic Human Environments

arXiv

Robust Dynamic Obstacle Avoidance with Optimistic-Pessimistic Model Predictive Control

HAL

Constrained Trajectory Optimization for Hybrid Dynamical Systems

Project page arXiv Code
Conference Presentations

ICRA 2026 — Vienna, Austria

  • Poster TD-CD-MPPI: Temporal-Difference Constraint-Discounted Model Predictive Path Integral Control
  • Workshop CACTO-BIC: Scalable Actor-Critic Learning via Biased Sampling and GPU-Accelerated Trajectory Optimization — Xplore Workshop

JNRH 2026 — Journées Nationales de la Robotique Humanoïde, Montpellier, France

  • Oral + Poster TD-CD-MPPI: Temporal-Difference Constraint-Discounted Model Predictive Path Integral Control
  • Tutorial TA Zero-Order Optimization Techniques for Robotics — teaching assistant for the half-day tutorial session
Service

Reviewer for ICRA, IEEE Robotics and Automation Letters (RA-L), IEEE Transactions on Robotics (T-RO), and IEEE-RAS Humanoids.

Research & Professional Experience

Oct 2024 – Present

PhD Researcher — Numerically Efficient Robot Learning

  • World models & model-based RL for sample-efficient policy learning and planning (CAST, with a sim-to-real Go2 handstand).
  • Universal policies that adapt in context (Transformer-XL) and control different morphologies with one network.
  • Learning + sampling-based planning with GPU-parallel MPPI (TD-CD-MPPI on Solo12), and distillation of planners into visuomotor policies (VAMPS).
AY 2026/27

Teaching Assistant — Robot Learning & Control

Tutor for Optimisation and Learning for Robot Control and Advanced Optimization-Based Robot Control (MSc Mechatronics, MSc AI Systems).

Jan – Jun 2024

Graduate Student Researcher

Fast DDP-based solver for constrained trajectory optimization of hybrid systems with contact, using exact derivatives and saltation matrices. MSc thesis work, published as arXiv:2410.22894.

Oct 2022 – Dec 2023

Motion Planning & Control Engineer

Algorithms for Autonomous Driving division of a Formula SAE driverless team. Trajectory planning and MPC-based tracking control for a fully autonomous race car.

Aug – Dec 2023

Junior Data Analyst

Climate risk assessment for Philip Morris International facilities worldwide. Geo-visualization pipelines for multi-model climate hazards across temperature scenarios up to 2050.

Education

Oct 2024 – Ongoing

PhD in Robotics & Optimal Control (Cotutelle)

Supervisors: Nicolas Mansard (LAAS-CNRS) and Andrea Del Prete (University of Trento). Numerically efficient RL for robotics, combining model-free and model-based methods with MPPI trajectory optimizers.

Sep 2022 – Jul 2024

MSc in Mechatronic Engineering

Specialization in Software Technologies for Automation. Erasmus+ year at Aalto University (GPA 5.0/5.0).
Thesis: Constrained Trajectory Optimization for Hybrid Dynamical Systems, a DDP-based solver for non-smooth dynamics with contact constraints. Supervised by Prof. M. Canale (PoliTo), co-supervised by Prof. V. Kyrki (Aalto) and Assist. Prof. G. Alcan (Tampere).

Sep 2019 – Jul 2022

BSc in Industrial Engineering

Specialization in Mechatronic Engineering.
Thesis: Automatic Differentiation: Theory and Applications in Nonlinear Programming, bridging CasADi and Autodiff with IPOPT. Supervised by Prof. E. Bertolazzi, co-supervised by Prof. F. Biral.

Skills

Languages Python · C/C++ · MATLAB · Julia
Frameworks JAX · PyTorch · ROS · CasADi · IPOPT · NumPy / SciPy
Simulation Isaac Lab · Isaac Sim · Isaac Gym · MuJoCo · MuJoCo MJX · MuJoCo Warp · Brax
Methods Model-based & model-free RL · world models · transformer policies (Transformer-XL) · imitation learning (ACT) · MPPI · MPC · DDP / iLQR
Robots Unitree H1 · Unitree R1 · Unitree Go2 · AlienGo · Solo12 · Flexiv Rizon 10S (RGB + F/T sensing)
Spoken Italian (native) · English (fluent) · French (basic)

Earlier Projects

Projects from my Bachelor's and Master's years, before starting the PhD.

  • MPC for Robotic Arm Control. Linearized MPC for a 4 and 5 DoF robotic arm, simulated and run on a physical model built by the team with 3D-printed links and servo motors, with David Costa (PoliTo) and Loïc Delineau (EPFL). Video report.
  • MPC for Formula SAE Racing Car Control. Linearized MPC for the racing car's longitudinal and transversal dynamics, reducing lap time under vehicle and environmental constraints. MATLAB, co-generating C code for onboard real-time performance.
  • C++ IPOPT Interface Using AD with CasADi. Part of the bachelor thesis on automatic differentiation. Algorithms for large sparse matrices, bridging CasADi and Autodiff with the IPOPT nonlinear optimization package.

Interests

Marathon Running Alpine Skiing Mountain Biking Climbing Trekking