Research

Autonomous decision-making for spacecraft.

I study reinforcement learning for space-based space surveillance and onboard tasking at CU Boulder's Autonomous Vehicle Systems Lab, advised by Hanspeter Schaub.

Space domain awareness

Space-based space surveillance

An orbiting optical sensor must track moving targets while managing its own attitude, power, storage, and communications. I develop closed-loop tasking policies that couple these spacecraft constraints with the geometry and illumination of resident space objects.

This research progresses from LEO-to-LEO imaging to observations across orbital regimes, then to persistent catalog maintenance with revisits and changing target priorities. The studies use Basilisk and BSK-RL spacecraft simulation.

Basilisk BSK-RL
Space-to-space imaging simulation from the AMOS 2026 presentation.
AMOS 2026 Figure 2 comparing LEO-only and mixed-regime training.
Training-catalog comparison · Fig. 2

AMOS
2026

Responsive Autonomous Resident Space Object Imaging and Catalog Maintenance Using Reinforcement Learning Across Orbital Regimes

Daniel Huterer Prats and Hanspeter Schaub

Persistent, priority-aware catalog maintenance by a single LEO observer imaging targets in LEO, MEO, and GEO. The formulation supports repeated observations, online target-priority promotions, and variable-duration imaging and downlink actions. The study adapts a target-wise attention policy and evaluates training-catalog composition, acquisition–delivery rewards, catalog-size transfer, and response to changing priorities.

Paper PDF Recorded talk Citation
Figures and presentation media
Figure 2: delivered ground value across reward weights for LEO-only and mixed-regime training.
Training-catalog comparison in matched mixed-regime simulations. Figure 2, PDF page 9.
Daniel presenting the AMOS 2026 research in Maui.
Presentation photos and simulation.
AMOS 2025 Figure 5 showing spacecraft resources and image delivery during an example mission.
Example mission resource history · Fig. 5

AMOS
2025

Reinforcement Learning for Space-to-Space Surveillance: Autonomous Scheduling for Resident Space Object Imaging

Daniel Huterer Prats, Hanspeter Schaub, and Chris Wheeler

LEO-to-LEO optical imaging formulated as a partially observable decision process, coupling dynamic target selection with onboard resource management. An actor–critic policy is trained in Basilisk and BSK-RL and compared with a heuristic scheduler under spacecraft, line-of-sight, and illumination constraints.

Paper PDF Recorded talk Citation Project and slides

Related journal manuscript: Journal of the Astronautical Sciences · Under review.

Figure and simulation
Figure 5: spacecraft resources, illumination, downlink access, and image delivery during one example rollout.
Resource history from one representative simulated mission. Figure 5, PDF page 11.
Polaris visualization from the AMOS 2025 presentation.

Ongoing research

Constellation tasking and communications

I am extending space-based imaging to decentralized multi-spacecraft catalog maintenance and tasking that accounts for communication quality. These projects are in progress.

Rendezvous and proximity operations

Autonomous satellite inspection

Close-range inspection requires a spacecraft to maneuver around an object, observe illuminated surfaces, and avoid collision. In co-authored work with Mark Stephenson and Hanspeter Schaub, we study closed-loop maneuver selection with an optimization-based safety shield.

Illustration of a spacecraft inspecting a nearby resident space object
IWPSS 2025 Figure 8 showing inspection fuel consumption, duration, and evaluation outcomes.
Shielded inspection evaluation · Fig. 8

IWPSS
2025

Autonomous Satellite Inspection in Low Earth Orbit with Optimization-Based Safety Guarantees

Mark Stephenson, Daniel Huterer Prats, and Hanspeter Schaub

A semi-Markov inspection formulation combines learned impulsive-maneuver decisions with an optimization-based collision-avoidance shield. The policy responds to the current relative spacecraft state rather than following a fixed offline maneuver sequence. The shield's guarantees are defined within the adopted Clohessy–Wiltshire–Hill relative-motion model.

Paper PDF Recorded talk Citation
Figure from the paper
Figure 8: inspection fuel consumption and duration, with evaluation outcome distributions.
Shielded inspection evaluation under the paper's relative-motion model. Figure 8, PDF page 8.

Earlier work

Trajectory design and astrophysics

Interplanetary trajectory with multiple planetary gravity assists

Master's project · UPC

Multi-gravity-assist trajectory optimization

I developed a genetic-algorithm approach to selecting gravity-assist sequences and transfer timing, comparing flight duration with total delta-v.

Project reportTrajectory-design reading notes
Figure from the dark matter and galaxy formation study

MNRAS 2021 · NYU Abu Dhabi

Dark-matter-deficient galaxies

My earlier research used galaxy simulations to examine tidal stripping. Our co-authored MNRAS paper studies how encounters with larger galaxies can leave dwarf galaxies with much less dark matter.

Journal articleOpen PDF