1st International Round Table on Intelligent Control for Space Missions

Friday 24 Nov 2017, 08:00 → 17:00 Europe/Amsterdam

Newton 2 (ESTEC)

Celia Yabar (Moltek for ESA), Dominika Perz (ESA), Guillermo Ortega (ESA), Jinesh Ramachandran (ESA), Maria Grulich (ESA)

Intelligent Control (IC) is a class of control techniques that use various artificial intelligence computing approaches like neural networks, Bayesian probability, fuzzy logic, machine learning, evolutionary computation and genetic algorithms.

The ICE (Intelligent Control for ESA) International Round Table has the following main objectives:

• Objective 1 (WHERE): Provide for a survey of the current state of the art in intelligent control applied in industrial process (transportation, manufacturing, personal computing) and space missions in particular.
• Objective 2 (HOW): Investigate and assess new, efficient,  and cost effective methods for the control of space systems by means of Intelligent Control IC techniques and technologies.
• Objective 3 (WHEN): To discuss the upcoming research and development opportunities for the ESA technology plans (TRP, GSTP) for the use of Intelligent Control IC in space missions.

The following topics are covered in the 1st ICE round table: Neural Networks control
Bayesian control, Fuzzy Logic control, Expert Systems and Artificial Intelligence, Genetic and Evolutionary control, lessons from Intelligent control for transportation systems, landscape of Intelligent control for manufacturing, personal assistants using intelligence, and Intelligent Control for space systems. Visit the Scientific Programme for more details.

Registration

Registration at the Newton registration desk

Conveners: Ms Celia Yabar (Moltek for ESA), Mrs Dominika Perz (ESA), Mrs Maria Grulich (ESA)

Session 1: Early Morning

Convener: Ms Celia Yabar (Moltek for ESA)

1 Introduction to the ICE round table

Introduction of the workshop, objectives, composition, structure, speakers. Keynote speech about the current ESA understanding of the state of the art and the needs for a more intelligent control in the upcoming complex space missions.

Speaker: Dr Guillermo Ortega (ESA)

Slides 1st_ICE_Introduction.pdf

2 "Space Missions Model-Based Control vs. Intelligent Control" by DLR

Modern model-based control methods such as model predictive control (MPC), inverse model feedforward control, and nonlinear dynamic inversion (NDI) encapsulate mathematical model representations of the system to be controlled into the feedforward or feedback control algorithm implementation. Depending on the accuracy and the validity range of the embedded model, these methods can achieve good performance and enable adaptation to the operating conditions of the physical system. During the workshop, a design process for physical model-based control methods based on the Modelica modelling technology will be discussed and applications in aerospace and robotics will be presented. The goal of this talk is to stimulate a discussion on how model-based and intelligent control could be utilized in a complementary or combined way for future space missions.

Speaker: Dr Johann Bals (DLR)

Slides ICE_2017.pdf

3 "AI Planning & Scheduling for a new generation of Mission Planning tools" by Solenix Deutschland GmbH

Artificial Intelligence has improved the cost-effectiveness and flexibility of mission planning tools design and development. This talk provides an overview of significant planning and scheduling experiences in deploying Intelligent tools to innovate the ESA mission planning practice. Specifically, the focus is in some key points that an approach based on Artificial Intelligence brings to Intelligent Control: the attention to domain modelling, the algorithm synthesis oriented to the explanability of the solution, the support for a mixed-initiative approach to the solving process, the relevance of user interaction services.

Speakers: Dr Nicola Policella (Solenix Deutschland GmbH), Dr Simone Fratini (Solenix Deutschland GmbH)

Slides AI4IC_Updated.pptx

4 "Fuzzy logic for the control of a CubeSat" by Technical University of Madrid

In this talk we show that the use of intelligent control systems based on fuzzy logic is a great advantage over conventional control systems currently being used in satellite attitude control, and open new possibilities of application of intelligent controllers in the field of space technologies. In order to develop and introduce intelligent systems in the space field, we have designed an adaptive fuzzy logic controller for a nanosatellite Attitude Determination and Control Subsystem (ADCS) and we have compared its performance and efficiency with a traditional Proportional Integrative Derivative (PID) controller. Fuzzy controllers have already been studied for satellite attitude control; however their performance has not been compared before with a classical PID controller typically being implemented on board spacecraft nowadays. In a first step, we have designed and implemented both controllers in order to test them on board a nanosatellite in the QB50 mission. Due to the requirements imposed by the payload, the orbit, and the significant power limitations in these small spacecraft, an efficient ADCS is required in order to fulfill the mission objectives. In a second step we have optimized numerically both controllers by using a genetic algorithm and we have compared the efficiency of the optimized controllers.

Speaker: Prof. María Victoria Lapuerta González (Technical University of Madrid)

Slides Fuzzy_logic_for_the_control_of_a_cubesat_-_VLG.pptx

5 "Identification of target from image by Deep Learning" by JAXA

This talk is dedicated to the efforts of JAXA in the field of artificial intelligence. And in particular about Intelligent Control using Deep learning as well as Reinforcement Learning and Machine Learning. The talk contains also the work done by JAXA on the areas of Onboard identification of target using Deep Learning and Machine Learning as well as in the field of Safety of Intelligent Control Systems.

Speaker: Keisuke Sugawara (Japan Aerospace Exploration Agency)

Slides Slide_Sugawara.pptx

Coffee Break 1

Session 2: Late Morning

Convener: Mrs Dominika Perz (ESA)

6 "Optimisation Technology for Intelligent Control" by University of Southampton

Optimisation is a key technology for intelligent control, as exemplified by evolutionary and genetic algorithms, machine learning techniques, and training of neural networks. Deterministic global optimisation is concerned with rigorous methods to find globally optimal solutions for such applications as well as mathematically precise error bounds for such solutions. Corresponding solver technology has made great strides over the last 40 years, and mixed-integer linear or quadratic problems with 100,000 variables or more can now routinely be solved to global optimality in practice. We provide a short survey of existing techniques and solvers as well as potential directions of further developments.

Speaker: Prof. Joerg Fliege (University of Southampton)

Slides globalopt_fliege.pdf

7 "Challenges and state-of-the-art of neural network verification" by fortiss GmbH

The deployment of Artificial Neural Networks (ANNs) in critical missions poses a number of new verification challenges. In particular, for ANN-enabled perception and control it is important to establish properties about the resilience of ANNs to noisy or even maliciously manipulated sensory input. Furthermore, given interpretability of the inputs and outputs of an ANN, certain safety properties can be verified. This talk summarizes the current state of the art of measuring the resilience and verifying safety properties of ANNs. We show the possibilities for application as well as future research challenges.

Speaker: Georg Nührenberg (fortiss GmbH)

Slides Challenges_and_state-of-the-art_of_neural_network_verification.pptx

8 "Applying Artificial Intelligence techniques to the orbit propagation problem" by University of La Rioja

Artificial Intelligence in general, and machine learning in particular, are the cornerstones of the so-called Industry 4.0. Being aware of the new opportunities provided by current technology and increased computational power, public administrations are fostering this upcoming fourth industrial revolution, in which both the private sector and academia are also taking an active part. The space sector, which is not an exception to this tendency, is looking for a similar advance to Industry 4.0. Among many other needs, improvement in the accuracy required for the solution of some practical Space Situational Awareness (SSA) problems is being actively sought. In particular, current needs in orbit determination and prediction, especially in the framework of SSA, are demanding innovative approaches. During the last years, we have been applying machine learning and statistical techniques to improve the accuracy of any kind of orbit propagator. For that purpose, we characterize the deviation of the propagator by means of the time series of its error with respect to observations or pseudo-observations, that is, accurately computed ephemerides. Then, we model that deviation through machine learning methods, such as neural networks or gradient boosting machines, and also by means of statistical methods. Finally, we are able to predict deviation values in the future, based on the developed model. In this talk, we will present this methodology, which we have called hybrid propagation, the basics of neural networks, the machine learning software that we use, the necessity to fine-tune hyper- parameters, the importance of developing parsimonious models, and the results of an illustrative example in which we correct the error of the well-known SGP4 propagator for Galileo orbits.

Speaker: Dr Juan Félix San-Juan (Scientific Computing Group (GRUCACI), University of La Rioja)

Slides RoundTable_JFSanJuan.pdf

9 "Deep Reinforcement Learning for Control" by the University of Stuttgart

Deep learning techniques allow us to scale reinforcement learning to problems that were previously intractable, i.e. to domains with high-dimensional state (or observation) spaces and continuous action spaces. We will give an overview of state-of-the-art deep reinforcement learning methods, including deep Q-learning, deep deterministic policy gradients, and asynchronous advantage actor-critic. We furthermore show the use of deep neural networks to approximate the optimal state-feedback control of continuous time, deterministic, non-linear systems - ranging from toy problems to fuel-optimal spacecraft landing. The deep neural control policies are obtained through imitation learning from optimal trajectories. The method is able to capture the optimal state-feedback with high accuracy and to generalize well beyond training data.

Speaker: Prof. Daniel Hennes (University of Stuttgart)

Slides deeprl_ustuttgart.pdf

Lunch

Session 3: Early Afternoon

Convener: Mrs Maria Grulich (ESA)

10 "Optimisation, Uncertainty Quantification and Data Analytic at the Intelligent Computational" by the University of Strathclyde

The Intelligent Computational Engineering Laboratory (ICE Lab) in the Department of Mechanical and Aerospace Engineering at the University of Strathclyde acts as the bridge between novel computational intelligence techniques and real-world engineering applications. The aim is two-fold: to learn from real-life problems and develop new solutions as well as to apply existing numerical techniques to explore new problem-solving strategies. During the workshop, an overview of the current activities in the field of evolutionary optimisation and mission analysis, evidence based robust optimisation and space system design, artificial intelligence and space traffic management will be presented.

Speaker: Dr Annalisa Riccardi (Strathclyde University)

Slides ICERoundTable_AR.pptx

11 "On-board intelligence for small space drones" by the University of Delft

Both in space and on earth there is a strong drive to make smaller autonomous robots. Smaller implies lower production costs and the possibility to scale up to large multi-robot systems. Specifically for space exploration, this might lead to a reduction in launching costs. A challenge is to make such robots still perform on a level that is comparable to larger robots, under strict restrictions of size, weight, and power (SWaP). During our presentation, we will highlight some of the frontiers in small autonomous robots, applied to both earth- and space drones. We will discuss recent advances in bio-inspired and constrained optimization approaches to creating extremely efficient artificial intelligence, self-learning drones (also in space), and autonomous flight of swarms, where drones have to avoid each other and coordinate with each other to achieve a global task. Finally, we will suggest directions of research that we deem very suitable for space application of the presented technologies.

Speakers: Dr Guido de Croon (TU Delft), Dr Javier Alonso-Mora (Delft University of Technology)

Slides Onboard_intelligence_small_drones_v2.pptx

12 "Applications of Intelligent Control in Industry and Adaption to Space Missions" by Knowtion

Currently, intelligent algorithms for optimization and control are employed in many so called Industrie 4.0/Internet of Things projects. These include on one hand the optimization of production processes by implementing intelligent control algorithms into production control or Manufacturing Execution Systems, and on the other hand the improvement of products by equipping them with intelligent algorithms. The project "Integrated Vehicle Health Management System Demonstragor (iHMSD)" was part of GSTP 4.1 and included a glider for high altitude re-entry. This talk gives an overview of possible approaches in various industries and explains the potential adaption of employed algorithms to iHMSD. Here, possible applications of intelligent algorithms for flight control and ground control of such a glider are presented

Speaker: Mr Vesa Klumpp (Knowtion UG)

Slides Vesa_Klumpp_Knowtion_2017-11-24.pptx

Coffee Break 2

Session 3: Late Afternoon

Convener: Mrs Maria Grulich (ESA)

13 "The expanding reach of Artificial Intelligence in Space Exploration" by JPL

NASA JPL has been employing Artificial Intelligence in support of space exploration. These efforts have been ongoing for many years: NASA’s Earth Observing One mission was controlled by AI software – the Autonomous Sciencecraft and Earth Observing Sensorweb – for more than a dozen years, linked with other satellites and scores of ground assets to track volcanism, floods, wildfires and the cryosphere. In the field of astronomy, machine learning is also serving to triage vast amounts of radio and optical data, enabling scientists to focus on the most relevant results. AI scheduling has also been playing a crucial support role for space operations – including overseeing the Deep Space Network, science operations and downlink/data management for JPL elements of ESA’s Rosetta comet-chaser and NASA’s Mars rovers – including the 2020 Mars Rover now in development. And as well as having crucial support roles on the ground, AI will also be leaving Earth: its role in future mission concepts to ocean worlds and, eventually, other stars will be highlighted.

Speaker: Dr Steve Chien (Jet Propulsion Laboratory, California Institute of Technology)

Slides ESTEC-Intelligent_Control_static.pdf

Round Table

Conveners: Mrs Dominika Perz (ESA), Mr Jinesh Ramachandran (ESA), Mrs Maria Grulich (ESA)