On the representation and awareness of context for underwater robots in marine environments

Tracking #: 1508-2720

Xin Li
Jose-Fernan Martinez
Gregorio Rubio

Responsible editor: 
Oscar Corcho

Submission type: 
Ontology Description
Enhanced context awareness is a necessity for underwater vehicles to behave intelligently and achieve potential coordination and cooperation. In order to present a complete picture of the marine environment for underwater vehicles, this paper presents an ontology to abstract heterogeneous contexts obtained from the marine environment and model their associated uncertainty based on the Bayesian Network (BN) theory. The proposed ontology is a networked ontology consisting of several modules, including users, sensors, vehicles, environmental context, probability, and external sources etc. It could formally represent heterogeneous contexts, integrate data from different sources, facilitate information reusing, and enable vehicles with context awareness. In addition, it could support multiple reasoning, including ontology, rule and BN based inference. An oil spill monitoring scenario is presented to verify the proposed ontology in terms of its extensibility, applicability, interoperability, reusability and capability of supporting multiple reasoning mechanisms.
Full PDF Version: 


Solicited Reviews:
Click to Expand/Collapse
Review #1
By Ghislain A. Atemezing1 submitted on 26/Dec/2016
Major Revision
Review Comment:

-- General statement --
This paper proposes an ontology to enhance awareness for underwater vehicles in marine environment. The proposed ontology uses NeOn methodology to build a network of modules consisting of different aspects that are relevant to the domain, such as sensors, users, context, uncertainty, etc. The authors evaluate the model against the oil spill monitoring scenario. A link to the ontology is available for evaluation.

-- Overall review --
Overall, the paper is easy to read and clear enough to understand the challenges of tackling both awareness and uncertainty in marine environment. The authors have made a significant review of the SoA in ontologies in similar domains. However, IMO, there are still missing convinced evidence of the applications of the ontology in its current state. I was disappointed to read in the conclusions that the ontology “has not been employed in real usages” while I was expecting this to have been tested in the project, and might have helped to adjust the ontology in a real world scenario with lessons learned.
The application domain of this work is really interesting. However what is missing is the clear evidence of the usage of the proposed ontology. What the authors mention in the future work as a step forward (the integration with the SWARMs platform), would have been clearly a real application of the proposed ontology.

-- Details review --
- Section 3: It’s not clear to the reader why do you choose the NeON methodology for this implementation, compare to any other approach, such as ontology patterns http://ontologydesignpatterns.org/. Was it easier to follow the methodology? What are the lessons learnt in this respect?
+ Section 4: Fig. 1 is clear enough as it shows the different layers of the modules. However it is missing an “integration module” that shows how to import/reuse piece of the modules. Could you give details about which OWL2 profile belong each of the module, together with the exact number of classes and properties. You mention that “a set of concepts are chosen from the marineTLO ontology”. How many concepts? It’s not clear from the text how many classes and properties of MarineTLO are finally used in the ontology.
- I am curious to understand the need to have a class “Unknown”. How is it relevant for your model since modeling with OWL is rather choosing the Open World Assumption by default?
- You define spatial relations such as “closeTo”, “adjacentTo” and “farFrom”. How do those properties related to the distance? When do you consider two seabed zones are far? 10m? 100m? Would it be better to define also rules based on distances and/or positions to specify those values? Also, regarding locations, how do you store the positions? just as point? (lat, long) or polygones?
- “the concept Fauna is aligned with the concept marinetlo:MarineAnimal”. Which type of alignement? owl:equivalentClass? Also, you mention the external sources without elaborating how they are imported. Are they instances? SKOS concepts? Just a remark: the FLOD data set here http://www.fao.org/figis/flod/ is 404.

+Section 5:
It is hard to find how you show to the reader the performance of the extended ontology. Let’s take the BN used for casual relationships. How is it integrated in the picture in Fig. 1?
You mention also that “the ontology is validated by marine experts and ontology engineers”? How was that validation took place? Make a brief statement on how that validation was made. What about the completeness of the ontology?
You mention the possibility to add user-defined rules, while there no mention of experts rules contained in the ontology. Do you mean there are only 2 SWRL rules for this extended ontology?
How do you combine in practice BN reasoning with OWL reasoning? This could be useful to understand better if you use two different reasoners and how you combine the results

-- Ontology resource review --
- There are 5 properties with multiple domain or range in the file “swarmsontology.owl”. According to OWL spec, they are allowed (https://www.w3.org/TR/owl-ref/#domain-def) but should be interpreted as conjunctions. However, it seems for some of them the authors want the use of disjunction (e.g., the property attachedTo), in that case, use owl:unionOf.
- There are multiple namespaces in the vocabulary (I see at least two, http://www.MarineContextOntology.orgggg# and http://www.MarineContextOntology.org ) that makes it difficult to understand if their provenance.
- There are missing labels (rdfs:label) and metadata to help reusing the ontology.
- It is not clear to me the axioms used to integrate the different modules upper-core-domain applications in the ontology file. Could you clarify that aspect in the paper?

-- Minor typos --
- In the text, you use so often “etc.”, especially in introduction and related work. Maybe reduce them for better readability.
- Add a space before a reference. e.g., “interests[21]” should be “interests [21]” and two more in section 2.
- s/S preadS peed/SpreadSpeed ; s/WindS peed/WindSpeed.

-- Suggestions --
-Sec 2: It would be great to also make all the vocabularies listed in this section available on the Web, so that the reviewers can also assess them.
- Please mention a link to each module of the ontology at the same time that you are describing in the text. Currently the link is mentioned almost in P.12 without specifying which module or if it is the entire ontology.
- Fig. 5 is not readable, at least in a printed version.
- P.12: Please rephrase this sentence “Given observed evidences …. “
- You might find some vocabularies for vehicles in LOV, like this list: http://lov.okfn.org/dataset/lov/terms?q=Vehicle

Review #2
By Ilaria Tiddi submitted on 27/Dec/2016
Major Revision
Review Comment:

The paper presents an ontology model to represent a marine environment with the main motivation that a modular, light-weight and reusable knowledge representation can facilitate the usage underwater autonomous vehicles. Although similar ontologies in the same area already exist, the novelty of the work is to provide a representation model that takes into account the heterogeneity of the underwater scenario and, more specifically, that considers the dynamism and uncertainty of the environment as key aspect. The result is an ontology including several interconnected modules (e.g. users, sensors, vehicles, environmental context, probability, external sources etc) organised in four layers (upper ontology, core ontology, domain ontologies, and application ontologies). After describing the main motivation and works at the state-of-the-art, the authors define the set of requirements (both from the literature and more domain-specific) that the ontology has to fulfil, then presenting a core description of each of the ontology modules. The applicability of the model is evaluated by instantiating the ontology in an oil spill detection use-case, and by extending it based on the specific needs of the scenario.

I like the work and I think it is well motivated, solid and clear. However, there are some things that could be clarified, which would help in improving the work. I am addressing them point by point below.

General remarks:
To my perspective, the most interesting thing that the paper suggests is that semantic web representations can be used to improve the efficiency of robotic tasks. My suggestion is to highlight this point much much more, especially in the introduction and motivation. Highlight that reasoning with heterogeneous knowledge is is still an unsolved challenge in robotics; say that SW technologies have been proven to be effective in managing knowledge from different sources but few works have suggested the integration of SW to improve the robot reasoning. You should strengthen your motivation: not only you are facilitating cooperation and context-awareness in underwater autonomous vehicles, but you are showing the SW community that their efforts are benefitting another area, proving the works in robotics that SW representations can be useful, and therefore encouraging communication/collaboration between these two communities.

Specific remarks, based on the requirements that the paper should meet:

(1) Quality and relevance of the described ontology
+ I think the ontology is of very good quality. The modularity allows reusability and extensibility, as shown in the section 5.
+ Each ontology layer is briefly but pointy described, and most of the design choices are motivated.
+ I appreciate that the authors chose to minimise the efforts to build the ontology and maximising the usage of existing vocabularies and ontologies.
+ The design principles have been clearly defined (in section 3) based on the existing literature, i.e. the Neon guidelines, and then extended with specific needs of the application domain. They seem reasonable, and have been respected when designing the ontology in section 4. For clarity, the authors might want to consider, when describing the modules of section 4, which of the requirement is met by each module.
+ A few more words could have been spent on the methodology used to build the model, e.g. giving a couple of sentences of what does the Methontology consists of, for non-expert readers? I am thinking particularly of readers that come more from the robotics field.
+ As for the relevance, convincing evidence of why such an ontology is needed is presented in Section 2, that shows both the lacks of the existing work and the novelty of the current one.
+ A proper comparative evaluation with other ontologies on the same topic is not explicitly provided. However, the differences between existing works and the presented ontology are clearly explained in Section 2. One suggestion to highlight more the strength of the work could be to explain, in the case study section, why the existing ontologies are not suited, and what are they missing.
+ A preliminary scenario of a use-case is presented in Section 5. Other scenarios, as well as more clear experiments, are mentioned as underway in the future work section. Could you also spend few words on how do you think the ontology could be evaluated? Could this based on different scenarios?
+ The ontology is free and publicly available.

(2) Illustration, clarity and readability of the describing paper
+ The paper is very well structured. The key aspects of the ontology are presented since the beginning of the work and are punctually repeated.
+ I hardly spotted English errors, and the paper is very nice to read. Some very minor remarks are given below.

Some points that I would like to have clarified:
+ your main point is that you need to include uncertainty because the underwater environment is subject to uncertainty and dynamism. You mention "partial views, data loss, and imperfect instruments, etc”. Is it not this the case of ground and aerial robots, too? My impression is that autonomous vehicles and robots are always subject to harsh conditions, namely because they operate in a real-world scenarios, which are highly dynamic. If you think this is not true, you should clarify more what makes the underwater environment more affected by uncertainty. From my understanding, the point of the paper is to design an ontology that contemplates uncertainty because existing ontologies cannot do it, so I would not speculate too much on it.

+ One thing I did not understand is, do you only consider object properties? Is there any specific reason why properties such as hasDepth, hasPosition, hasPressure, hasSalinity, whose range should be values, are chosen to be object properties? Do you have a predefined set of possible values? I have seen that you have some datatype properties though, so which is the difference?

+ I am not an expert of fuzzy reasoning but I was wondering, what is the reason of choosing the Probability model rather that a fuzzy representation? I know this is mentioned in the conclusions as future work, but more motivation should be given of why that model is preferred. Any pros or cons? Is it because of the Bayesian reasoning?

Very minor stylish things:
+ remove the ~ in front of your \footnote{…}, since it creates a blank space between the word and the apex, e.g. robot~\footnote{myrobot} > robot\footnote{myrobot}
+ change the \titlerunning{} with your actual title, at the moment you have “Instructions for the preparation of a camera-ready paper in LATEX”
+ page 2, section 1 : "Conclusions are given in section 6 and future work is pointed out as well." is weird, reformulate
+ page 3, section 3.1 : "To present a complete picture of the marine environment for vehicles and operators, any context that is relevant to the application domain and the environment is the modelling of interest" I think it should be reformulated too. I have problems in finding the part of speech here
+ your usage of "including" is wrong, I think. English uses "including" as "not excepting" ("we are all going to the cinema, including the children" https://en.oxforddictionaries.com/definition/including), but my impression is that you use it to say "consisting of". To avoid confusion I would replace it with “consisting of", e.g. "requirements, including non-functional and functional” > "requirements, consisting of non-functional and functional"
+ this is probably very picky, but should the properties not be in 3rd person (singular)? e.g. contains, provides, observes etc?
+ page 9, change caption of Figure 4 (which is a copy of Fig. 3)
+ page 11, to improve readability of Figure 6 and 7, please consider other RDF serialisations, e.g. N3, and namespaces
+ page 13, section 6 "an ontology with its aim to represent" > "an ontology to represent / aiming at representing"
+ page 13, section6 "underwater vehicles under the context of the SWARMs project" > "underwater vehicles in the context of the SWARMs project"
+ page 13, section 6, first item of the list in the right column, remove the second "."

Review #3
Anonymous submitted on 06/Mar/2017
Major Revision
Review Comment:

This manuscript was submitted as 'Ontology Description' and should be reviewed along the following dimensions: (1) Quality and relevance of the described ontology (convincing evidence must be provided). (2) Illustration, clarity and readability of the describing paper, which shall convey to the reader the key aspects of the described ontology.

The paper makes a thorough and utterly sound description of the modelling process for an ontology for the representation and awareness of context for underwater robots in marine environments. Interestingly the ontology includes probability-extended context representation and copes with spatial relationships, integrating it with well-known relevant ontologies both outside and within the domain modelled. Although a use case is shown to demostrate the modelling capabilities of the solution, no effort is done in order to test the scalability of the ontology modelled. The work should be extended to assess not only the expresiveness of the solution, but also the impact that this expressiveness might have on the additional load in terms of storage and computing due to the usage of rich and powerful semantic modelling. Rules that make use of the probabilistic modelling of the solution should be shown. Performance results in terms of a realistic knowledge and rule base to model a realistic underwater surveillance case should be provided. A more detailed example to further support the expresiveness validation is advisable. The ontology is relevant and well-argued. The paper contents are clear, but the value of the resulting ontology has to be further demonstrated with an extended evaluation, including not only expressiveness but also scalability aspects. A few English mistakes identified. Perform a careful extra review of the text.