Dr. Dario Izzo

Head of ESA's Advanced Concepts Team

Bio

Dr. Dario Izzo is Head of ESA's Advanced Concepts Team, the European Space Agency's internal scientific think tank dedicated to blue-sky research and disruptive innovation in space science and technology.

A graduate in Aeronautical Engineering from Sapienza University of Rome (1999), he further specialized in Satellite Platforms at Cranfield University (UK) before completing his Ph.D. in Mathematical Modelling at Sapienza, where he assisted Prof. Chiara Valente in courses on rational mechanics and spaceflight mechanics.

He is widely credited as a pioneer of AI techniques applied to spaceflight mechanics, with seminal contributions that predate and, in many ways, anticipated the current AI revolution. Among his most influential innovations are G&CNets (Guidance & Control Networks), which demonstrated that neural networks could autonomously perform real-time spacecraft guidance and control; GeodesyNets, which introduced neural implicit representations for the modelling of irregular celestial body shapes and gravity fields; and ThermoNets, bringing neural computation to bear on the modelling of Earth Thermosphere. To date, he has authored over 230 scientific papers at the intersection of space engineering and machine intelligence.

His influence extends well beyond the academic literature. Dr. Izzo created the Global Trajectory Optimisation Competition (GTOC), now an iconic benchmark challenge engaging hundreds of researchers worldwide, as well as a suite of follow-on open problem initiatives, Kelvins, Optimise, and SPoC, that continue to push the frontier of what machines and humans can solve together.

He is the recipient of the Humies Gold Medal, awarded for the creation of an AI system demonstrating human-competitive performance in the design of interplanetary trajectories, one of the most prestigious award in evolutionary and genetic computations.

Abstract

Modern astrodynamics is undergoing a fundamental transformation, driven not by new hardware but by a new mathematical language, one in which neural computation and classical mechanics speak the same grammar. This talk presents some of the original ideas that are behind this transformation, with efficient differentiability and Taylor integration as one of their computational backbone.

At the foundation lies a single enabling insight: when neural networks are integrated directly into the ODE formalism of dynamical systems, their flows can be expanded analytically to high order on differential event manifolds. This yields Event Transition Tensors (ETTs), a rigorous tensorial description of how uncertainties, perturbations, and adversarial inputs propagate through neural dynamical systems, without resort to Monte Carlo sampling or linearization. ETTs provide, for the first time, a pathway to certifiable and interpretable neural dynamics, with explicit mathematical structure replacing the black-box opacity typically ascribed to deep networks.

This framework is the theoretical capstone unifying many landmark neural architectures, among which G&CNets encode optimal guidance and control policies onboard spacecraft, learning the deep structure of Pontryagin's optimality principles across interplanetary transfers, planetary landings, and proximity operations. GeodesyNets exploit neural implicit representations to reconstruct the gravity fields and irregular shapes of small bodies from sparse observations. ThermoNets deploy neural surrogates for spacecraft thermal modelling in complex radiation environments. In each case, the same differentiable infrastructure — efficient automatic differentiation coupled to Taylor integration — enables both the training and the rigorous verification of these networks.

The talk explores how ETTs open genuinely new scientific questions at the intersection of Hamiltonian mechanics and machine learning: from learning governing equations of chaotic three-body systems, to characterizing trajectory uncertainty in neurally controlled landing scenarios near binary asteroids — directly relevant to ESA's HERA mission.

Zheng H. Zhu

Professor and Tier I York Research Chair in Space Robotics and Artificial Intelligence Department of Mechanical Engineering York University

Bio

Dr. Zheng Hong (George) Zhu is the Editor-in-Chief of Acta Astronautica and a Professor & Tier 1 York Research Chair in Space Robotics and Artificial Intelligence in the Department of Mechanical Engineering at York University in Canada. His research spans spacecraft dynamics and control, tethered space systems, autonomous space robotics, computational control methodologies, and in-space additive manufacturing. He has published over 250 peer-reviewed journal papers and 190 conference papers.

Dr. Zhu is an academician of the International Academy of Astronautics; a College Member of the Royal Society of Canada; a Fellow of the Canadian Academy of Canada, the Engineering Institute of Canada, the Canadian Society of Mechanical Engineering (CSME), and the American Society of Mechanical Engineers (ASME). He is also the Associate fellow of the American Institute of Aeronautics and Astronautics (AIAA).

His achievements have been recognized with numerous prestigious awards, including the 2024 Solid Mechanics Medal and 2021 Robert W. Angus Medal (CSME), the 2024 Gold Medal and 2019 Engineering R&D Medal of Ontario Professional Engineers Awards, and the 2021 York University President’s Research Excellence Award.

Abstract

This work introduces a new approach reinforcement learning approach for motion planning of a 6-degree-of-freedom (DOF) free-floating space robotic manipulator, focusing on challenges like pose alignment and obstacle avoidance, especially in dynamic environments with coupling to the spacecraft base. The approach is trained in a multi-scenario environment that includes random initial conditions and obstacles to ensure robustness across a wide range of potential missions. A unique reward function encourages precise end-effector alignment, smooth joint motion, and effective obstacle avoidance, while using quaternions to prevent singularities and enhance resilience to observation noise. After training in simulation, the method is validated on a hardware-in-the-loop testbed, showing promising results. This research provides a strong foundation for integrating reinforcement learning into future space robotic operations, paving the way for more autonomous and efficient space missions.

Dr. Steve Chien

Jet Propulsion Laboratory, California Institute of Technology

Bio

Steve Chien is a Technical Fellow in Artificial Intelligence and Co-head of the Artificial Intelligence Group at the Jet Propulsion Laboratory, California Institute of Technology. He has spent decades deploying AI/Autonomy to numerous space missions including: Earth Observing One, Sensorweb, ESA’s Rosetta Orbiter, and M2020. He has been awarded five NASA Medals in 1997, 2000, 2007, 2015, and 2025 for development and deployment of AI technologies for space missions. He has supported numerous government bodies including the Defense Science Board and the Air Force Scientific Advisory Board. He was appointed by Congress to the National Security Commission on Artificial Intelligence (2018-2021). He also served on the Army Science Board from 2023-2025.

Abstract

In October 2023, the Onboard Planner (OBP) took control of the Perseverance rover on Mars, over 200 million miles from Earth. As of December 2025, OBP has operated for over 400 tactical plans covering over 700 Martian days (sols) and has: executed over 12000 activities requested by scientists and engineers, driven over 20 kilometers, acquired over 100,000 images, and collected more than 10 rock core samples. In contrast to the traditional form of operations, where operators provide a rigid set of instructions for the rover, with OBP Perseverance revises its schedule an average of 16 times each day to stay responsive in a dynamic Martian environment where things don’t always go as expected. This flexibility allows the mission to manage resources such as energy more efficiently and therefore accomplish more science. 

In this talk, we discuss the approach to ensuring that a search-based AI system, specifically the Onboard Planner, would (1) achieve mission objectives; and critically (2) protect the rover, a multi-billion dollar one of a kind asset. We describe the “whole lifecycle” approach to developing trusted autonomy software for M2020, spanning: conception, design, analysis, prototyping, and testing. We then describe the incremental rollout and training to smooth the transition to operations with increased onboard autonomy. Next we discuss how the OBP software has improved mission return in quantity and quality in several ways. Finally we describe the even greater challenges of autonomy in future missions to hunt for life beyond Earth.

Simone D’Amico 

Associate Professor, Stanford Aero/Astro 

Chief Scientist and Co-Founder, EraDrive Inc. 

Bio

Simone D’Amico is Associate Professor of Aeronautics and Astronautics and Geophysics (by Courtesy), Science Fellow at the Hoover Institution, Co-Founder and Chief Science Officer at EraDrive Inc. He is the Founding Director of the Stanford Space Rendezvous Laboratory, Founding Co-Director of the Center for AEroSpace Autonomy Research (CAESAR), and Director of the Undergraduate Program in Aerospace Engineering at Stanford. He has 23+ years of experience in research and development of autonomous spacecraft and distributed space systems. He developed and deployed the distributed Guidance, Navigation, and Control (GNC) system of several formation-flying, swarming, rendezvous and proximity operations missions (e.g. PRISMA, TanDEM-X, and BIROS). Currently, he is the institutional PI of four autonomous satellite swarms funded by NASA (STARLING, STARI) and by NSF (VISORS, SWARM-EX). Dr. D’Amico is Fellow of AIAA and AAS. He is in the Advisory Board of four space start-ups focusing on distributed space systems for future applications in SAR remote sensing, orbital lifetime prolongation, and space-based solar power. He was the recipient of several awards, most recently the 2024 NASA Ames Honor Award and the 2025 SpaceNews Icon Award for the Starling mission, Best Paper Awards at AMOS (2025), IAF (2022), IEEE (2021), AIAA (2021), AAS (2019) conferences, and the M. Barry Carlton Award by IEEE (2020). He received the B.S. and M.S. degrees from Politecnico di Milano (2003) and the Ph.D. degree from Delft University of Technology (2010). 

Abstract

Distributed space systems are transforming spaceflight by replacing single monolithic spacecraft with multiple coordinated assets such as formations, swarms, and servicing vehicles that can observe, assemble, inspect, and maneuver cooperatively. Yet these missions raise the bar for onboard autonomy in order to be technically feasible and financially viable: spacecraft must perceive, decide, and act safely in uncertain, dynamic, and resource-constrained environments, often with limited ground intervention. This keynote will discuss frontier research on how to achieve these objectives through the judicious integration of artificial intelligence or learning-based approaches into the spacecraft autonomy stack to enhance rather than replace optimization-based guidance, navigation, and control. The talk will highlight the most recent advances obtained at Stanford in leveraging multi-modal transformers for vision-based perception and navigation for unknown non-cooperative targets, semantically consistent trajectory generation for rendezvous and proximity operations, and autonomous reasoning for spacecraft guidance. Ultimately, this work marks an evolutionary step toward vision-based, language-conditioned, constraint-aware and optimal spacecraft trajectory generation, enabling operators to shape both mission intent and safety-critical behavior through intuitive natural-language commands while reducing expert burden across future science, commercial, and defense missions.