Project ElectroKarst

Objectives

Previous work done in the scope of REK initiative put in light three main robotic challenges that the present proposal, ElectroKarst, proposes to address:

Objective 1: Develop a highly maneuverable and safe robot, called in the sequel ElectroKarst Explorer (EKE), for a confined and unstructured flooded environment. The proposal is to design and engineer a polyarticulated robotic system of anguilliform type, and equipped with thrusters. The advantages will be to obtain a compact system able to evolve in very narrow environment, with varying actuation geometry that can face strong current (in case of a reduction of the karst section that produces an increase in the water current) or provide full and high maneuverability when station keeping is required.

Objective 2: To generate a reactive behavior that naturally leads the robot as far as possible from the walls and to penetrate autonomously as far as possible into the karst labyrinth. As the clarity of the water in this underground context cannot be guaranteed, vision navigation cannot be a reliable single solution. Moreover, the confined nature of the environment imposes strong constraints on the parameterization of acoustic sensors (multipath and potential interference between acoustic devices). The proposal is therefore to implement the principle of "electric sense" on the poly-articulated system and to adapt its use to the karst context. It is worth noting that the slender profile of the carrier is particularly well suited to the electric sense (Boyer et al, 2013), while variations in its geometry will offer new possibilities for improving the electric sense abilities.

Objective 3: Perform an efficient localization in this GPS-denied environment, developing a SLAM algorithm based on acoustic measurements and electrical sense measurements with a set-membership approach. In this context where no obvious landmark exists, we need to build a sensor-based map which allows us to eliminate some inconsistent pairs of poses which correspond to a loop closure. This strategy already applied to bathymetric SLAM (Rohou et al, 2019) can be applied to any situations where an unstructured environment (such as a Karst) should be used for the localization. On top of that, it allows state estimation even if the observation function is not completely known, which can be the case when dealing with electrical sensing. The guarantees carried out by the set-membership approach allows to perform an on-line and reliable localization performance monitoring and adapt the mission parameters accordingly. This approach will allow us to trig the deployment of heavy magnetic localization devices at the surface in order to proceed to calibration when necessary.

Objective 4: Perform experimental validation tests on field with increasing-complexity procedures, demonstrating the utility of this new system and providing useful data to the institutional partners that are in charge of the sites management. Pertinent sites have already been identified. They have been chosen according to their geo-morphology, accessibility and their societal importance. The concurrent realization of these objectives will demonstrate a new type of system specifically suited to autonomously visit inaccessible confined environments. The main application is the groundwater resource assessment, which presents true contemporary societal, technological and scientific challenges.

Position of the project as it relates to state of the art

A rapid historical view of the use of robots in karstic exploration: The first documented trials on the use of robotic systems in karst environments date from the 1960s, when Cdt Cousteau and H. Delauze used the Télénaute for an exploration of the Fontaine de Vaucluse, and initiated a series of expeditions whose successive failures were rich in lessons: the robots were successively lost, the last ones coming to be caught in the remains of the umbilical cables of the previous ones. Thus, the question of the management of the umbilical cable linking the robots to the surface is essential in this confined and unstructured environment. One solution may be to do without it. This is the approach chosen by the 3 current projects whose scientific and applicative objectives are close to the REK initiative: UnexMin and its successor UnexUp and DepthX. The first two projects propose to develop an autonomous system to inspect old flooded mines to decide whether to reopen them. This is a structured environment, without current, and in which the water is supposed to remain clear. DepthX is built to perform an acoustic mapping of geological artifacts such as cenotes, gigantic drowned vertical chasms. The vertical morphology of cenotes allows for an autonomous robotic solution, due to the simplicity of the trajectory and the direct and vertical acoustic communication that is possible between the machine and the supervision system, at surface. Thus, contemporary initiatives opt for an autonomous solution that imposes conditions either on the environment - its structure and the clarity of the water - or on the trajectories that the system can achieve. The karst context does not allow such assumptions and the originality of the REK initiative lies in its exploration strategy which consists in breaking down the mission into two phases:

i. The exploration phase, during which the robot explores an unknown environment and in which it is essential that the expert (hydrogeologist) can control the progress of the mission and indicate the zones of interest. The robot is equipped with a trancanneur (or lost optic fiber) which allows the system to lay down its umbilical cable as it penetrates in the karstic conduit. Its onboard instrumentation (mainly acoustic to remain robust to the turbidity of the water) allows it to draw up a geomorphological model (3D) of the of the visited environment. ii. The return phase, initiated by the operator or by an unexpected break of the cable, during which the robot separates from the umbilical cable and performs the reverse path fully autonomously, using the map built during the previous phase for an acoustic SLAM-type navigation.

ii. The return phase, initiated by the operator or by an unexpected break of the cable, during which the robot separates from the umbilical cable and performs the reverse path fully autonomously, using the map built during the previous phase for an acoustic SLAM-type navigation.

The first experimental realizations of the REK initiative were carried out with various projects: REEA, ALEYIN and LEZ 2020. These preliminary studies have addressed the following topics :
• On-board instrumentation adapted to the karst environment: study and implementation of new types of sensors: principles of the 'acoustic skin' (Alarab, 2020) and the 'active cable' (Troesch, 2022).
• Redundant actuation with variable geometry and active redundancy management: the usefulness of implementing more motors than necessary to acquire a fine control of the robot’ reactivity (Ropars, 2018) and proposal of a mechatronic structure for dynamic adaptation of the actuation geometry (Dang, 2020).
• Graph-SLAM using acoustic sensors in karst environment (Breux, 2020).
• Seaking for a guaranteed mapping using an intervallist approach.
• Non-singular control of the attitude of an underwater robot in a confined environment (Lapierre, 2021).
• Logistical management and field experimentation: essential aspect of the project in relation with the composition of the team implementing the experiments (speleodivers, roboticists, hydrogeologists...), in situ. Expeditions carried out in the chasm of Gourneyras, the resurgence of Durzon, the fountain of Saint-Antoine, the fountain of Nîmes, and the springs of LEZ.

Based on this past experience, the limitations that emerged led to the elaboration of the needs expressed by the objectives 1, 2 and 3 of this project.

Poly-articulated underwater systems:
Bio-inspiration has produced a lot of interesting designs for underwater systems. From eel-type (McIsaac, 2003 – Khalil, 2007) to tuna fish (Wolfgang, 1999), the idea of investigating new type of locomotion using natural solutions inspiration has conquered scientific mind all over the world. But, for the moment and from the energetic performance point of view, no one has shown a more efficient solution than the use of a rotating propeller, underwater. Meanwhile, another interesting property of a poly-articulated serial system is its ability of changing its shape and its actuation configuration when propellers are mounted on it. The Eelume system (Liljebäck, 2017) follows the same idea and proposes a compact and highly maneuverable robot, dedicated for deep-sea infrastructures inspection and maintenance. The novelty of our design will be to adapt the necessary sensor’s suite for karstic exploration, optimize the actuation system configuration to actively manage the maneuverability, force attainability, reactivity and energy consumption according to the environmental conditions and constraints (Fig. 3). Moreover, as explained here after, the adjunction of the electric-sense principle will also raise new constraints on system shape. Finally, the control architecture design of the whole system is raising original and exciting scientific questions on the control of the redundant actuation system (composed with configuration articulations and thrusters) that couples actuation efficiency and observation capability.

SLAM in confined and underwater condition:
The simultaneous localization and mapping (SLAM) problem (Leonard,1992) for an autonomous robot moving in an unknown environment is to build a map of this environment while simultaneously using this map to compute its location. The history and critical issues of SLAM are discussed in (Frese, 2006). SLAM methods can be classified in two categories (Thrun, 2005), which are referred to as feature-based SLAM and location-based SLAM. Feature-based SLAM assumes that the map is composed of a set of features together with their Cartesian location. The map has, thus, a parametric structure where the features are points, segments, corners, or any other parametric shape (Castellanos, 2001). The feature-based SLAM problem can be cast into a state-estimation problem by including the feature parameters among the state variables (Montemerlo, 2003). Probabilistic techniques (Kalman filtering, Bayesian estimation, and particle filters) or set membership approaches have been proven to efficiently solve the feature-based SLAM problem (Marco, 2001). Now, feature-based maps are not well suited to model non-structured environments, as it is the case for underwater robotics in a karstic environment where landmarks have no particular geometric shape. Location-based maps offer a label to any location in the world. They contain information not only about obstacles in the environment but about the absence of obstacles as well. They can also map what should be sensed at a given location of the robot. A classical location-based map representation is known as occupancy map (Elfes, 1987) (also called pose-based map). They assign to each point of the world an occupancy value (a Boolean number or a probability of occupancy) that specifies whether or not a pose is occupied by an obstacle. This concept can be extended to assign to each point the value that should be returned by the sensor for each pose of the environment. This extension is particularly suited to our underwater context where acoustic or electric sensors can hardly be interpreted in terms of geometrical constraints. Moreover, the map uncertainty which is needed in our karstic world can easily be interpreted in terms of a thick set (Jaulin 2011).

New sensorial approach for confined environment
The confined and unstructured nature of the karst environment places a heavy burden on sensor performance. Environments characterized by large amounts of clutter present significant challenges (Thorpe, 2001), particularly due to the difficulties in using acoustic methods (Knight, 1981). Acoustic devices must be finely tuned to match frequency and amplitude to local environmental conditions to avoid multipath, reverberation and attenuation effects. Difficulties continue to increase in turbid waters, in which vision is of little use. Therefore, a wide range of application situations remain beyond the capabilities of existing technology. Because of these inherent physical limitations, robotic underwater navigation in confined spaces and turbid waters still remains a challenge. To address this, one promising prospect is to draw inspiration from the several hundred species of freshwater fish that have developed an original sense, called the active electrical sense. Discovered by Lisman and Machin in the 1950s (Lissmann & Machin, 1958), the electrical sense allows these fish to navigate, detect prey and predators, and communicate with each other in the turbid, obstacle-saturated waters of African and South American rainforests. Because of these environments, the electrical sense is a short-range omnidirectional sense. When using the active electric sense, a weakly electric fish emits an electric field in its near environment whose refractions by obstacles are measured by its electro-sensitive skin and interpreted in its brain (von der Emde & Schwarz, 2002 - Pereira et al, 2012). Another modality of electrical sense is used by the same fish (Hopkins, 2005), and by some saltwater fish such as sharks and rays (Kalmijn, 1966). It is called passive electric sense, because in this case the fish directly detects electric fields emitted by any organism living in the water. In the case of weakly electric fish, many are gregarious and have evolved complex social behaviors such as those involved in their courtship displays that are primarily supported by their ability to communicate with their electric fields (Gebhardt et al, 2012). In its active modality, the range of electrical sense is about one fish length, while it is about two to three times longer in the passive mode. Due to the lack of sensor technologies for underwater robotics in confined spaces and turbid waters, this bioinspired sense, now called artificial electric sense, has been applied to several underwater robotics problems ranging from reactive navigation (Boyer et al, 2013, 2015), object localization (Silverman et al, 2012 - Lebastard et al, 2010) and recognition (Bai et al, 2012 - Lanneau et al, 2017), or teleoperation using haptic feedback (Fang et al, 2016 - Boyer et al, 2019). All of these results have been achieved with novel bio-inspired sensors measuring electric fields by voltages (Solberg et al, 2007) or currents (Servagent et al, 2013). Remarkably, the sensors developed by LS2N (Servagent et al, 2013) reproduce the range of fish in both active and passive sensing.

Equipped with these new algorithmic and mechatronic solutions, we will study the performance of the ElectroKarst Explorer in real conditions. In order to do so, we have identified different experimental sites in which we will follow an experimental roadmap with increasing complexity. In order to do so, we will use an intermediary experimentator, the system NavScoot (NS, Fig. 3), an underwater scooter on which the robot’s sensors are mounted. This allows for the acquisition of real data in real conditions. Then, once the algorithms are calibrated on real data, the sensors and the control architecture will be deployed on the robotic system, the ElectroKarst Explorer and tested in real conditions.

Research projects : ElectroKarst

Objectives

Position of the project as it relates to state of the art

Partners

Robot

Reports