Giusy Falcone, PhD

×

Autonomous Aerobraking Implemented via PR-DRL

Aerobraking enables a cost-effective method to insert a spacecraft in a stable orbit around a planet. However, it requires long mission time, costly operations, and ground station coverage and is highly susceptible to human decisions and atmospheric density variability. Conventionally, a spacecraft performing an aerobraking drag passage is required to transit in a heat rate feasible corridor, representing the region in which the constraints of the mission are achieved, and its safety is guaranteed. Currently, the navigation team is in charge of predicting the following periapsis locations and performing propulsive maneuvers if the predicted periapsis is outside the corridor. Performing autonomous aerobraking on board would decrease the operational costs and free the mission from human decision capability. This study develops a more efficient decision-making process via deep reinforcement learning.

To tackle the challenge of autonomous aerobraking planning, a parallel domain randomized deep q-learning architecture is developed. This architecture has a double function to speed up the learning process for heavy simulation-based application and to improve generalization.

The PR-DRL architecture is used to develop a planning capability for aerobraking planning maneuvers at the apoapsis that relocate the periapsis to avoid thermal violations during atmospheric passages and target a final orbit. Specifically, the Markov-Decision process framework is applied to the 2001 Mars Odyssey aerobraking campaign. A three-dimensional running reward function expressed in terms of spacecraft state and action was designed. The proposed algorithm was trained for the 2001 Mars Odyssey mission from the Main Phase 2 to the Walkout Phases. Specifically, the latter is the most hazardous phase of the entire mission because the spacecraft would shortly impact the planet's surface if the apoapsis decay is not strictly controlled. Historically, closer to the planet surface and more the heat rate was constrained; however, our approach shows that an onboard decision capability would enable enough control capabilities to relax these constraints. Results show that the PR-DRL approach outperforms the current state-of-the-art heuristic, achieving an average reward 130% higher than the achieved heuristic median score. Results also show that the spacecraft can proactively and robustly avoid thermal violations and develop a behavior that allows reaching the target orbit 70% of the time in a broadly perturbed environment.

Papers

[2022] Falcone, G., Putnam, Z. R. (2022). Autonomous Decision-Making for Aerobraking via Parallel Randomized Deep Reinforcement Learning. IEEE Transactions on Aerospace and Electronic Systems.

[2022] Falcone, G., Putnam, Z.R. (2022). Deep Reinforcement Learning for Autonomous Aerobraking Maneuver Planning. In 2022 AIAA SciTech Forum. San Diego, CA. (2022 AIAA Intelligent Systems Best Paper Award)

Invited Talk

[2018] Falcone, G. (2018) ML for Autonomous Aerobraking In 15th Annual International Planetary Probe Workshop IPPW-2018. Boulder, CO

×

Aerobraking Trajectory Control using Articulated Solar Panels

Aerobraking is a process used to insert a spacecraft in a low orbit around a planet with limited fuel consumption. The maneuver is composed of many passages in the upper atmosphere of the planet, which slow down the spacecraft and decrease its orbital energy. The solar panels represent not only the largest drag area in aerobraking and thus the most liable surface to slow down the vehicle, but also the most exposed surface to the thermal environment.
This study proposes a new and redesigned idea of aerobraking in which the solar panels are used to provide the in-plane control authority to the spacecraft during the atmospheric passage. The use of articulated solar panels transforms the spacecraft into a body subjected to a variable drag, which with a continuous rotation of the solar panels can control the atmospheric trajectory.

This new concept of operation has the advantage of being able to (a) quickly compensate for the density variations with the rotation of the main drag area and (b) control the thermal conditions. In addition, the solar panels' rotation also enables mission designers to (c) target a final state at the atmospheric exit. An onboard guidance algorithm was implemented to control the atmospheric passage by rotating the solar panels. The control algorithm exploits the real-time density and temperature data (IMU and thermocouples data) to maximize the dissipated energy during an atmospheric passage while maintaining an acceptable thermal environment. Specifically, optimal control theory was used to implement the onboard guidance algorithm. The optimal control solution states that a max-min-max solution, where the max value change during the drag passage, is necessary to maintain the heat rate and heat load around the defined limit. An example of a solution is presented in the Figure below (left). The guidance algorithm runs online and uses the drag passage approximate solution to define the control switches.

Results show that in a mission similar to that of Mars Odyssey, the use of rotating solar panels allows a reduction of approximately 79% in the number of required atmospheric passages and of approximate 76% in the time of flight (from 52 to 12 days) to complete the entire aerobraking campaign.

Papers

[2021] Falcone, G., Putnam, Z. R. (2021). Energy Depletion Guidance for Aerobraking Atmospheric Passes. AIAA Journal of Guidance, Control, and Dynamics, 56(6), 1689-1703.

[2019] Falcone, G., & Putnam, Z. R. (2019). Aerobraking Trajectory Control Using Articulated Solar Panels. In 2019 Astrodynamics Specialist Conference . Portland, ME.

×

Closed-Form Trajectory Solution For Shallow, High-Altitude Atmospheric Flight

This study looks for a closed-form approximate solution for shallow, high-altitude atmospheric flight. Currently, a solution that includes expressions for velocity, flight-path angle, and altitude for lifting, high-speed atmospheric flight has been already published, with a complete derivation of it. The current study solution is based on the assumptions of small flight-path angles and altitude rate changing linearly with respect to time. Comparison between direct integration of the equations of motion and closed-form solution is presented in the figure below proving that the found formulas achieve a good match with simulative results.

However, the current study required the definition of an ε term, which is held accountable for the energy depletion in the atmospheric passage and is directly related to the flight-path angle. The current follow-up of this study attempts to generalize the definition of ε and to relax the initial assumptions. In the video below, I present the current findings of this study. The video was recorded for the 2020 AAS/AIAA Astrodynamics Specialist Conference.

Papers

[2020] Falcone, G., & Putnam, Z. R. (2020). Closed-Form Trajectory Solution For Shallow, High-Altitude Atmospheric Flight. In 2020 Astrodynamics Specialist Conference . Lake Tahoe, CA.

×

Aerobraking Trajectory Simulator - Open Source on GitHub!

The Aerobraking Trajectory Simulator (ABTS) is a Python-based modeling and simulation tool developed to assess flight performance for the aerobraking mission set. The tool is open-source for the benefit of the entire aerospace community. ABTS can be used for research purposes in mission design and analysis for the aerobraking mission set. The simulator can be used to model an entire aerobraking campaign, a single orbit, or a single atmospheric passage around Earth, Mars, and Venus, while providing estimates of important flight performance parameters, including dynamic pressure, heat rate, and energy depletion rate. The drag passage environment can also be used to simulate entry trajectories. The computational efficiency of the tool also enables uncertainty analysis through Monte Carlo techniques. The built-in modularity of the tool enables selection of physical models from a range of built-in options, including a method to efficiently incorporate complex atmospheric models, such as Mars-GRAM 2010.

Results indicate ABTS is able to accurately model complete aerobraking campaigns, as shown by direct comparison to Mars Odyssey aerobraking data; additionally, it is also able to simulate reentry flight with blunted-cone, as shown through a Mars Pathfinder-inspired mission simulation.

Papers

[2021] Falcone, G., Putnam, Z.R. (2021). Design and Development of an Aerobraking Trajectory Simulation Tool. In 2021 AIAA SciTech Forum. Nashville, TN.

×

Drag-Modulation Aerocapture Maneuver at Mars

The increasing ability of smallsats to perform science missions at relatively low cost has led to the interest in using them as rideshare on interplanetary missions. Packaging and mass constraints limit the application of onboard propulsion systems on rideshare missions, leading to a limited range of orbits available to these interplanetary smallsat missions. The use of aerocapture drastically increases the range of orbits available to smallsat missions by decreasing their required propellant mass fraction. A relatively simple method of aerocapture is to use drag modulation to control the energy of the atmospheric exit state.

The study focuses on four baseline missions: a low-altitude Mars mapping orbit, Phobos and Deimos flyby/rendezvous, and areosynchronous orbit. Analyses include assessment of the sensitivity of the entry corridor size over different atmospheric conditions, a comparison of velocity-trigger and numerical predictor-corrector guidance schemes for drag modulation, and aerocapture flight performance assessment via Monte Carlo techniques.

Both the numeric predictor-corrector (NPC) and velocity trigger aerocapture guidance algorithm options provide adequate performance, though NPC is more accurate. Additionally, at low altitudes, the velocity trigger encounters an unacceptable amount of failed insertions, making it infeasible for these missions. A 5000 km of target radius guarantees to be far enough to complete successfully an aerocapture maneuver using the velocity trigger guidance algorithm.
The improved accuracy of the NPC method results in a Δv savings between approximately 75 and 155 m∕s when compared with the velocity trigger. Results indicate that the orbit insertion mass fraction for an aerocapture system may be as little as 33% of that for a fully propulsive orbit insertion system.

Overall, the NPC algorithm is able to deliver dry mass fractions between 60 and 75% to a variety of orbits in the Mars system.

Papers

Falcone, G., Williams, J. W., & Putnam, Z. R. (2019). Assessment of Aerocapture for Orbit Insertion of Small Satellites at Mars. Journal of Spacecraft and Rockets, 56(6), 1689-1703.

Falcone, G., Williams, J. W., & Putnam, Z. R. (2018). Aerocapture System Options for Delivery of Small Satellites to Mars. In 41st Annual AAS Rocky Mountain Section Guidance and Control Conference, 2018 (pp. 271-284). Univelt Inc

×

Chariot to the Moons of Mars

In this project, partnered with Purdue University, Arizona State University, Tyvak Inc., and JPL Team Xc, we developed a small satellite mission concept to investigate Phobos and Deimos. The project, also, aimed to demonstrate the drag-modulated aerocapture for orbit insertion purposes. I worked on the trajectory design pre- and post- aerocapture maneuver, the joint-plan for post-aerocapture maneuver to minimize the phasing time to initialize the moons rendezvous, and the aerocapture corridor definition.

×

Control for the Asteroid Redirect Mission (CIF Project JPL)

NASA’s Asteroid Redirect Robotic Mission (ARRM) aimed to pick up a boulder from a large asteroid and transport it to a distant retrograde orbit around the Moon for future exploration by a manned mission.

In this project, the team worked on a detailed analysis for one of the main control challenges in ARRM, i.e., three-axis attitude control of the ARRM spacecraft with the captured boulder in the presence of large uncertainties in the physical model of the boulder. After the modeling of the 30 degree-of-freedom nonlinear dynamic model of the ARRM spacecraft and boulder combination, the team linearized it about the nominal operating conditions to study the system’s modal properties. A finite element model of the ARRM spacecraft and boulder combination was used to validate our model. We then studied linear and nonlinear control laws for the attitude control problem. The proportional-derivative linear controller with lead-lag compensation and roll-off filtering is designed using standard design techniques. The robust nonlinear tracking control law guarantees global exponential convergence of tracking errors with finite gain Lp stability in the presence of modeling uncertainties and disturbances, and it reduces the resultant disturbance torque. The robustness properties of the control laws are demonstrated through simulations of the closed-loop system and computation of generalized gain and phase margins. Both control laws were capable of stabilizing the system despite uncertainties in the physical parameters of the boulder for standard scenarios, but only the nonlinear controller was suitable for more challenging scenarios involving large slew angles maneuvers along more than one axis.

Papers

[2016] Falcone, G., Saxena, A., Bandyopadhyay, S., Chung, S.J. and Hadaegh, F., (2016). Attitude control of the asteroid redirect robotic mission spacecraft with a captured boulder. AIAA/AAS Astrodynamics Specialist Conference (p. 5645)