Adam Evans
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Adam Evans has a total of 10 co-authors. Here are the top co-authors by number of shared publications:
- Roberto Armellin
University of Auckland
6 shared publications - Laura Pirovano
University of Auckland
5 shared publications Claudio Bombardelli
1 shared publication- Claudio Bombardelli
Technical University of Madrid
1 shared publication Harry Holt
1 shared publication- Harry Holt
European Space Agency
1 shared publication
Latest Publications
- Open Access
Sample-Free Safety Assessment of Neural Network Controllers via Taylor Methods
In recent years, artificial neural networks have been increasingly studied as feedback controllers for guidance problems. While effective in complex scenarios, they lack the verification guarantees found in classical guidance policies. Their black-box nature creates significant concerns regarding trustworthiness, limiting their adoption in safety-critical spaceflight applications. This work addresses this gap by developing a method to assess the safety of a trained neural network feedback controller via automatic domain splitting and polynomial bounding. The methodology involves embedding the trained neural network into the system's dynamical equations, rendering the closed-loop system autonomous. The system flow is then approximated by high-order Taylor polynomials, which are subsequently manipulated to construct polynomial maps that project state uncertainties onto an event manifold. Automatic domain splitting ensures the polynomials are accurate over their relevant subdomains, whilst also allowing an extensive state-space to be analysed efficiently. Utilising polynomial bounding techniques, the resulting event values may be rigorously constrained and analysed within individual subdomains, thereby establishing bounds on the range of possible closed-loop outcomes from using such neural network controllers and supporting safety assessment and informed operational decision-making in real-world missions.
Open AccessSample-Free Safety Assessment of Neural Network Controllers via Taylor Methods
In recent years, artificial neural networks have been increasingly studied as feedback controllers for guidance problems. While effective in complex scenarios, they lack the verification guarantees found in classical guidance policies. Their black-box nature creates significant concerns regarding trustworthiness, limiting their adoption in safety-critical spaceflight applications. This work addresses this gap by developing a method to assess the safety of a trained neural network feedback controller via automatic domain splitting and polynomial bounding. The methodology involves embedding the trained neural network into the system's dynamical equations, rendering the closed-loop system autonomous. The system flow is then approximated by high-order Taylor polynomials, which are subsequently manipulated to construct polynomial maps that project state uncertainties onto an event manifold. Automatic domain splitting ensures the polynomials are accurate over their relevant subdomains, whilst also allowing an extensive state-space to be analysed efficiently. Utilising polynomial bounding techniques, the resulting event values may be rigorously constrained and analysed within individual subdomains, thereby establishing bounds on the range of possible closed-loop outcomes from using such neural network controllers and supporting safety assessment and informed operational decision-making in real-world missions.
- 6 February 2026
- Published by Elsevier in Acta Astronautica
- Vol. 244, 1-19
Learning-based optimal guidance for spacecraft close-proximity operations with certified stability
- 6 February 2026
- Published by Elsevier in Acta Astronautica
- Vol. 244, 1-19
Learning-based optimal guidance for spacecraft close-proximity operations with certified stability
- 1 August 2025
- Published by American Institute of Aeronautics and Astronautics (AIAA) in Journal of Guidance, Control, and Dynamics
- Vol. 48 (8) , 1962-1972
Analytical Steering Law for Rapid Low-Thrust Fuel-Constrained Reachability Computation
- 1 August 2025
- Published by American Institute of Aeronautics and Astronautics (AIAA) in Journal of Guidance, Control, and Dynamics
- Vol. 48 (8) , 1962-1972
Analytical Steering Law for Rapid Low-Thrust Fuel-Constrained Reachability Computation
- 1 December 2024
- Published by American Institute of Aeronautics and Astronautics (AIAA) in Journal of Guidance, Control, and Dynamics
- Vol. 47 (12) , 2512-2524
Low-Thrust Fuel-Optimal Guidance with Automatic Control Sequence Detection and Separation
This paper develops a novel fuel-optimal guidance scheme using differential algebraic techniques that is capable of handling changes in the optimal thrust sequence, in addition to a large domain of uncertainty. Taylor polynomials of the solution flow exhibit large approximation errors when the control sequence changes within the domain of operation. This property is leveraged through automatic domain splitting, a technique that tracks the approximation error and splits the domain into smaller subdomains when high errors are observed. After multiple splits, the domain is autonomously separated along the boundaries between different optimal control sequences. Crucially, each resulting subdomain encompasses only a single control sequence, ensuring that any guidance updates obtained from the final polynomial map provide the corresponding optimal control sequence. To ensure high accuracy of the guidance updates, this work also develops an efficient and iterative-less differential-algebra-based refinement algorithm, which is applied to the outputs of the polynomial map. The overall guidance scheme provides highly accurate guidance updates in response to large trajectory deviations, including new optimal control sequences, as demonstrated in an Earth–Psyche transfer.
- 1 December 2024
- Published by American Institute of Aeronautics and Astronautics (AIAA) in Journal of Guidance, Control, and Dynamics
- Vol. 47 (12) , 2512-2524
Low-Thrust Fuel-Optimal Guidance with Automatic Control Sequence Detection and Separation
This paper develops a novel fuel-optimal guidance scheme using differential algebraic techniques that is capable of handling changes in the optimal thrust sequence, in addition to a large domain of uncertainty. Taylor polynomials of the solution flow exhibit large approximation errors when the control sequence changes within the domain of operation. This property is leveraged through automatic domain splitting, a technique that tracks the approximation error and splits the domain into smaller subdomains when high errors are observed. After multiple splits, the domain is autonomously separated along the boundaries between different optimal control sequences. Crucially, each resulting subdomain encompasses only a single control sequence, ensuring that any guidance updates obtained from the final polynomial map provide the corresponding optimal control sequence. To ensure high accuracy of the guidance updates, this work also develops an efficient and iterative-less differential-algebra-based refinement algorithm, which is applied to the outputs of the polynomial map. The overall guidance scheme provides highly accurate guidance updates in response to large trajectory deviations, including new optimal control sequences, as demonstrated in an Earth–Psyche transfer.
- Open Access
- 24 November 2024
- Published by Elsevier in Acta Astronautica
- Vol. 228, 17-29
Fuel-optimal guidance using costate supervised learning with local refinement
- 24 November 2024
- Published by Elsevier in Acta Astronautica
- Vol. 228, 17-29
Open AccessFuel-optimal guidance using costate supervised learning with local refinement
- 1 February 2024
- Published by American Institute of Aeronautics and Astronautics (AIAA) in Journal of Guidance, Control, and Dynamics
- Vol. 47 (2) , 279-290
High-Order Guidance for Time-Optimal Low-Thrust Trajectories with Accuracy Control
This paper develops a high-order guidance scheme based on the expansion of the solution to a time-optimal low-thrust optimal control problem by using differential algebraic techniques. The result allows new optimal trajectories to be autonomously computed from the simple evaluation of polynomials, including updated times of flight. Furthermore, a differential algebraic technique known as automatic domain splitting is implemented to improve the guidance scheme by adaptively and automatically splitting the uncertainty domain into smaller subdomains, in which new guidance polynomials are then generated which are better equipped to handle deviations in their respective subdomains. The result is not only a more effective handling of deviations in the trajectory compared to previous algorithms which use just a single set of polynomials but one that can provide the control and time-of-flight updates to a desired accuracy across the entire domain. Combined with its high-order nature, the guidance scheme is capable of dealing with nonlinearity and large deviations with respect to the reference trajectory, thus being robust to mismodeling and unplanned events. The guidance scheme is evaluated in three scenarios: simplified rendezvous using Clohessy–Wiltshire equations to demonstrate automatic domain splitting in the guidance problem, an Earth–Mars transfer, and an asteroid landing trajectory. The results show the ability of the differential algebraic-based guidance to provide optimal control and time-of-flight updates and improvements of several orders of magnitude over previous differential algebraic-based schemes in handling large uncertainty experienced in spaceflight by leveraging automatic domain splitting.
- 1 February 2024
- Published by American Institute of Aeronautics and Astronautics (AIAA) in Journal of Guidance, Control, and Dynamics
- Vol. 47 (2) , 279-290
High-Order Guidance for Time-Optimal Low-Thrust Trajectories with Accuracy Control
This paper develops a high-order guidance scheme based on the expansion of the solution to a time-optimal low-thrust optimal control problem by using differential algebraic techniques. The result allows new optimal trajectories to be autonomously computed from the simple evaluation of polynomials, including updated times of flight. Furthermore, a differential algebraic technique known as automatic domain splitting is implemented to improve the guidance scheme by adaptively and automatically splitting the uncertainty domain into smaller subdomains, in which new guidance polynomials are then generated which are better equipped to handle deviations in their respective subdomains. The result is not only a more effective handling of deviations in the trajectory compared to previous algorithms which use just a single set of polynomials but one that can provide the control and time-of-flight updates to a desired accuracy across the entire domain. Combined with its high-order nature, the guidance scheme is capable of dealing with nonlinearity and large deviations with respect to the reference trajectory, thus being robust to mismodeling and unplanned events. The guidance scheme is evaluated in three scenarios: simplified rendezvous using Clohessy–Wiltshire equations to demonstrate automatic domain splitting in the guidance problem, an Earth–Mars transfer, and an asteroid landing trajectory. The results show the ability of the differential algebraic-based guidance to provide optimal control and time-of-flight updates and improvements of several orders of magnitude over previous differential algebraic-based schemes in handling large uncertainty experienced in spaceflight by leveraging automatic domain splitting.
- Open Access
- 8 January 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in TechRxiv
Analytical Steering Law for Rapid Low-Thrust Reachability and Feasibility Computation
The reachable and feasible sets of spacecraft are important tools for many areas of astrodynamics such as mission design, space situational awareness, and assessment of potential threats. Whilst these may be obtained by solving many optimal control problems in the indirect formulation, this process is slow and can suffer from convergence issues with a lack of a good initial guess. This work presents an analytical method of rapidly estimating the reachable and feasible sets utilizing the D matrix, an error state transition matrix under linearity assumptions for Keplerian motion which employs time as a state variable and true anomaly as the reference variable. Analytical expressions for the control laws are derived to produce the maximum/minimum change in orbital radius, time-offlight, and out-of-plane height. Moreover, an analytical expression for the control efficiency is derived to allow thrust activation only when it is efficient to do so, allowing a target fuel consumption to be attained. By mixing the contributions of these controls, the complete reachable or feasible sets can then be rapidly swept over with minimal effort, which is demonstrated in a series of numerical simulations. The results show that the optimal control problem solution presents only minor improvements, highlighting the capacity of the D matrix to provide marginally sub-optimal results within a fraction of the computational time.
- 8 January 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in TechRxiv
Open AccessAnalytical Steering Law for Rapid Low-Thrust Reachability and Feasibility Computation
The reachable and feasible sets of spacecraft are important tools for many areas of astrodynamics such as mission design, space situational awareness, and assessment of potential threats. Whilst these may be obtained by solving many optimal control problems in the indirect formulation, this process is slow and can suffer from convergence issues with a lack of a good initial guess. This work presents an analytical method of rapidly estimating the reachable and feasible sets utilizing the D matrix, an error state transition matrix under linearity assumptions for Keplerian motion which employs time as a state variable and true anomaly as the reference variable. Analytical expressions for the control laws are derived to produce the maximum/minimum change in orbital radius, time-offlight, and out-of-plane height. Moreover, an analytical expression for the control efficiency is derived to allow thrust activation only when it is efficient to do so, allowing a target fuel consumption to be attained. By mixing the contributions of these controls, the complete reachable or feasible sets can then be rapidly swept over with minimal effort, which is demonstrated in a series of numerical simulations. The results show that the optimal control problem solution presents only minor improvements, highlighting the capacity of the D matrix to provide marginally sub-optimal results within a fraction of the computational time.