• No results found

I. LITERATURE REVIEW, OBJECTIVES AND METHODOLOCGY

3. Problem Definition, Objectives, Hypotheses and Methodology

3.1. Problem Definition

This research project focuses on the following problem: the IMU-only measurements cannot answer to the need for highly accurate navigation of the future lunar landing missions. The position and the velocity accuracies of such navigation system are respectively in the order of 50 km and 5 m/s. Some mission scenarios need a horizontal position accuracy of 100 m (3𝜎) as well as a horizontal and vertical velocity accuracy of 0.25 m/s (3𝜎) at landing [21].

The use of optical measurements, particularity the absolute ones, is then a promising avenue to fulfill these challenging requirements. However, vision-based absolute navigation techniques are not flight ready and many questions are still open for research. As a matter of fact, the candidate has identified many problems from the literature review presented in Chapter 2. These problems are related to the mission scenarios, to the image processing software, to the vision-based state-estimation filter and to the system validation. Each of these problems is summarized in the following paragraphs.

3.1.1.

Mission Scenarios Problems

Many aspects of the mission scenario for lunar pin-point landing (landing with a subkilometer horizontal landing accuracy) are still open design issues. In fact, the minimum set of sensors to fulfill the previously-cited navigation requirements has to be established. More precisely, the necessity of altimeter and/or star-tracker in addition of the vision measurements must be addressed. Furthermore, the time frame in which the sensors are used during the mission is not yet determined. In addition of having a significant impact on the navigation accuracy, the sensor usage sequence influences other aspects of the mission. For instance, the time frame in which the absolute optical measurements are used will determine the amount of memory required to store the geo-referenced

feature database (the features of a large part of the lunar surface would need to be stored on-board if the absolute optical measurements are used during the entire landing sequence). Furthermore, the camera embedded on the spacecraft has to be defined in terms of lens characteristics, image sensor characteristics as well as its mounting orientation on the spacecraft. The purpose here is not to design a particular mission or a particular camera, but to define the parameter space that influences the design and to identify the critical areas where the design margins are constrained.

3.1.2.

Image Processing Software Problems

The robustness of the image processing algorithm is an important issue that must be addressed. In fact, the global-feature techniques are robust to illumination changes, but are sensitive to noise and to the characteristics of the terrain (geographic phenomena similar such as craters, old crater edges and craters inside another). In contrast, the local-feature techniques are sensitive to illumination changes between the camera images and the geo-referenced map, but they can be used over a larger variety of terrains. In addition, features matching algorithms that are robust to spacecraft altitude and attitude are not yet fully developed and validated in a representative environment. The accuracy of the feature detection is another aspect where there is room for improvements. It is critical for the mission success since it has a direct impact on the vehicle state estimation accuracy. Finally, most of the image processing algorithms are still too complex to implement them in a space mission and require high computational power (typically not available on-board spacecraft). The consequences are that all of the state-of-the art approaches are not ready to be used in a high-reliability and high- cost space mission.

3.1.3.

Vision-Based State Estimation Problems

The vision-based state estimator also raises many questions. Despite the fact that many authors propose to use an Extended Kalman Filter (EKF), the best choice for navigation filter has to be consolidated and justified. The computational complexity of the relative optical measurement update must be addressed and it is not clear what could be the best method to fuse absolute optical measurements. In addition, the most efficient way to deal with the measurement delays must be determined. The estimator must be able to fuse many kinds of sensors using a robust and flexible architecture. In fact, many implementations of vision-based estimator require an augmentation of the state vector for the only purpose of the state update with the optical measurements. This complicates the integration of the technology to already-developed and well-proven inertial navigation systems. The architecture must support the addition or the withdrawal of sensors

according to the needs of the mission without a complete redesign of the navigation software. It must also be able to efficiently deal with sensors running at various rates, with sensors that are enabled only during a given time interval during the mission and with sensor failures.

3.1.4.

Navigation System Validation Problems

The validation and the performance assessment of vision-based navigation algorithms are very challenging issues. On the one hand, numerical simulations are not fully able to demonstrate the performance of the navigation system in a representative environment. In fact, complex phenomena are often difficult to model with a sufficient level of accuracy. For instance, the generation of realistic camera images based on the surface topographic, the surface albedo, the illumination conditions, the lens characteristics and the image sensor characteristics are very complex to simulate. In addition, many models require an enormous computational burden preventing the execution of simulations in a reasonable amount of time. On the other hand, the full scale-demonstration using an Earth-analog spacecraft is too expensive. The validation of vision-based navigation system is a critical element to resolve in order to propose their use for the next planetary exploration missions. At the time these lines were written and to the best knowledge of the candidate, no publication presenting the end-to- end and high-fidelity validation of a vision-based navigation system with synthetic image generation is available. On top of that, no real-time validation of absolute navigation with hardware in the loop has been performed.

Related documents