Project Karst

What's karst ?

Figure (1) : Classic Karstic Topography, from [1].
Karst generally comprises a network of underground natural conduits, resulting from the dissolution of soluble rocks, limestone, dolomite and gypsum, which may drain groundwater on a large scale. In karst aquifers, which supply drinking water to millions of people worldwide, these conduits correspond to preferential groundwater flow paths. In karst area, the access to the groundwater resource is generally constrained by the knowledge of the underneath conduit network since wells drilled into karst aquifers for water supply must intersect these conduits. In such areas, it is thus of major importance to get information about the geometry of theses preferential flow paths. This is a major and urgent, issue for public authorities concerned by the prospection, protection and management of the groundwater resource in karst regions. Assessing the geometry of flow paths network in karst, which constrain the dynamics of groundwater and transport processes, is an ambitious scientific objective that requires field information, which may be difficult to acquire. Cave diver is heroic, but faces physiological limitations. The use of a robotic solution may induce a significant evolution, in its capacity to go further and deeper in the karst maze. This requires an interdisciplinary scientific journey where hydrogeologists, mathematicians, electronic and control scientists share the same objective. This transdiciplinary posture is necessary to achieve the RKE's objectives. Moreover, the confined and chaotic conditions impose to keep the expert in the system’s decision loop during the exploration phase. He is indeed the best to decide on the system’s global and opportunistic objectives. This requires a communication link capable of streaming the current data acquisition, acoustic, or visual if turbidity allows. In the underwater environment, where wireless communication has very poor quality (bandwidth, latency), an umbilical cable is mandatory. Nevertheless, this cable is a heavy burden that is not admissible for the way back. Hence, the system has to be able to get rid of its cable, and return back autonomously. This question of varying autonomy is one of the exciting scientific issues on which the RKE project proposes to progress. As it will be exposed in the sequel, RKE presents a true and complete challenge, in terms of interdisciplinary academic research, engineering and socio-economical impact.

Scientific and technological challenges

Figure (2) : Global strategy, co-controlled way-in, autonomous return. .
Clearly robotic solution will have a significant impact on both objectives, karst exploration and gallery inspection; if satisfying solutions to actual technological and scientific locks are found. The overall ambition can be described by sensors system with controlled mobility. Indeed, the final objective is to bring back quality data from human inaccessible location. But in this demanding confined context, the robot cannot be seen as a ‘simple’ sensor carrier. The environmental data has to be used to perform safe and accurate navigation. Hence, sensors information has to be treated on line, providing environment models in order to perform real- time SLAM based navigation, a real challenge considering the constrained computational capabilities of the system. Moreover, pertinent (doubtlessly high-level) information has to be provided to the expert, coping with some a priori or guess knowledge, experimental protocols, analysis structure... Ideally, the robotic system should not add experimental complexity, and ‘disappear’ from the chain between the environment sampling and the expert needs. This scientific posture raises a lot of interdisciplinary processes, clearly and deeply impacting both the system architecture design – User (knowledge) Oriented Architecture [Lasbouygues 15] – and new available protocols for the expert. The confined environment requires a reactive system [Ropars 15], able to insure its own safety. This reactive behaviour can be achieved with proper control architecture and appropriate sensors. During exploration, where no a priori knowledge is available, the expert presence in the control loop is mandatory, but conditioned by the quality of the link between operator station and system. The umbilical cable provides high bandwidth communication, but clutters the system on the way back. Hence, we define 2 different phases of the mission: the exploration phase, where the system has to be able to autonomously lay down its umbilical cable, and the return phase where the system has to be able to get rid of its cable and perform the homing autonomously. This strategy indeed implies other system requirements, which will be exposed in the sequel.

Umbilical Management : in both situations (karst and gallery) expert needs to be in the loop. Indeed, the terrain specialist is the best resource to analyse and conduct exploration or inspection missions. Since underwater acoustic provides a very low bandwidth, the presence of the umbilical cable is mandatory. Nevertheless, it is not realistic to imagine a ROV system able to drag the umbilical cable along kilometres of chaotic relief (karst) or even regular structure (gallery). Hence, an onboard motorized secable truncanner is necessary, but is also a complex and delicate mechatronic device which requires a particular attention. Moreover, the presence of this umbilical, as reported by previous karst exploration attempts, is the major cause of failure since the cable is highly subject to remain blocked within the relief of the environment, specially during the return phase. In this phase, it has to be noticed that the cable, even unplugged is equipping the environment, and could be advantageously used as the diver’s lifeline. This implies to develop an active cable that guides the homing navigation.

Control and Co-control : the full teleoperation of a ROV system is already a difficult task for the operator in an opened environment, and requires hours of training. In the confined context, the task is even more difficult and an efficient control solution cannot rely on full teleoperation. Hence, the envisaged solution relies on co-control, where the system autonomously performs the control of given degrees of freedom, while the remaining ones stay under the operator control. In the karst exploration, or the gallery inspection, this distributed control decision allows for an autonomous centring of the system within the confined environment, while the system’s progression along the karst development is left to the operator’s decision. Several functioning mode have to be defined, in function of the system phases and the operator’s objectives.

- During exploration phase, where the umbilical is connected :

Co-controlled linear progression: the system is autonomously centred, using acoustic or video sensors, insuring its own safety, while the operator controls the system’s progression and attitude (orientations), as exposed in [Lasbouygues 14]. This safe autonomous centring requires reactive control architecture and new sensors development. Indeed, commercial profiling sonars exhibits very poor quality in terms of sampling period, while they offer range and precision for 3D mapping. Hence, reactive control is attainable with these classes of sensors at the cost of a very low forward progression or heavy assumptions on the environment morphological disparity. New type of sensor is necessary, requiring a co-design with the control. RKE proposes to develop the principle of an acoustic skin, providing the necessary information to feed the safe centring reactive control architecture.
Safe localized observation: this mode allows the operator to drive the system closer to the environment, in order to precisely observe a region of interest. Once again, the system has to autonomously guarantee its safety with respect to the environment, but with a different objective than the previous mode. In this context, observation sensors (acoustic or video cameras, curentmeter, thermometer...) can be advantageously used for local reactive navigation.

- Return phase, without umbilical

Autonomous homing: once the umbilical is cut, the system switches to decisional autonomy. This necessarily impacts the control architecture and this question of varying autonomy is quite challenging. During the return phase, the autonomous system should benefit from the knowledge acquired during the previous exploration phase. This knowledge is a collection of geo-referenced features extracted from the raw sensors data of the exploration phase (note that the computational burden of such treatment will be reasonably allocated to the operator station, as long as umbilical is plugged). The challenge here is to transform a heavy set of raw data into a collection of pertinent models to perform a SLAM based navigation, compatible with an on-line exploitation. This question mixes the statistical approach of SLAM techniques with geometrical compression via features extraction, i.e. stochastic mapping and multi-modal modelling.
Guarantees of Performance: the autonomous condition of the system, especially in such a harsh environment, imposes to address specifically the 5 different axes of the autonomy, as proposed in [Crestani 14]: stability, safety, energy, localisation and duration. The transposition of this approach to karst exploration will allow reifying a precious indicator relatively to the mission success, a ‘risk-taking’ estimator, as cave- divers do. This approach affords the system with versatility, fault tolerance and reliability.

Figure (3) :new envisaged sensors/technologies required for exploring a subaquatic confined environment. .
New sensors development : The environmental condition in which the system evolves imposes to adopt a new posture on the sensor question, and the resulting environment sampling. As mentioned before, commercial products do not afford the specifications, neither on the control requirements nor the expert needs. Commercial devices are made for offshore applications, where systems are generally over-dimensioned and energetically supplied by the umbilical cable. Our application requires lighter systems, in terms of weight, size and energetic consumption. Moreover, from the expert point-of-view, long-range karst mapping is more a question of volume estimation and geo-referenced approximate geometry estimation, than centimetric 3D modelling. On the other hand, safe reactive control requires precise and high-rate measurements, over a restricted range. The acoustic skin is a proposition to solve this issue. The active umbilical is another innovation that RKE proposes to realise.

Navigation (global) : The system geo-referencing within the karst maze is a hard problem. As in underwater condition, GPS information is available only at surface. If some solutions exist for submarines, using surface craft to relay satellite information with USBL 21 techniques, solution for confined environment is still an active research topic. Let’s mention the work of the ISSKA22 institute who develops the first underground GPS, based on a portable magnetic loop and 4 receivers at surface. The main restriction of this system is its size, weight and its range, over 100m23. But the interest remains and contact has been established with ISSKA to initiate collaboration on this topic. Nevertheless, we have to consider that direct geo-referencing is not available. Hence, the global navigation has to be performed using dead-reckoning and SLAM techniques on a priori knowledge. Note that acoustic positioning is still an option, relatively to a ‘distance’ measured between the operator station and the system. During exploration phase, the length of the laid cable can also provide an estimation of this distance. We can also reduce the dead reckoning drift with sensors correlation, or optical (acoustic?) flow analysis. An important requirement is to exploit a priori knowledge. We might have a rough topography made by divers, where characteristic regions have been identified (syphons, particular geometry...) and (approximately) located. Note that this a priori knowledge is certainly worst than the system can do. But during the exploration phase, the goal is to avoid loosing the robot, and any information, even of poor quality, is precious. This is obviously a very challenging SLAM problem.

Multi-modal stochastic mapping and SLAM : The exploration phase results in a huge collection of data that has to be treated in order to provide a pertinent environmental model to help system localisation during the return phase. This data treatment is a complex issue that has to cope with specific constraints as i) preservation of characteristic regions, amers and ii) exploitability within a SLAM approach coping with onboard computation power in order to be included in the control loop. Statistical approach and model reduction are the main tools to achieve this objective.

Research projects : Robots for Karst Exploration

What's karst ?

Scientific and technological challenges

Partners

Test sites and XP

France

Portugal

Spain

Italy

Germany

Croatia

Robotic Systems

Master Reports

Research projects : Robots for Karst Exploration

What's karst ?

Scientific and technological challenges

Partners

Test sites and XP

France

Portugal

Spain

Italy

Germany

Croatia

Robotic Systems

Related Projects

Master Reports